Deep-learning-enabled multi-omics analyses for prediction of future metastasis in cancer

Abstract

Metastasis remains the leading cause of cancer-related mortality, yet predicting future metastasis is a major clinical challenge due to the lack of validated biomarkers and effective assessment methods. Here, we present EmitGCL, a deep-learning framework that accurately predicts future metastasis and its corresponding biomarkers. Based on a comprehensive benchmarking comparison, EmitGCL outperforms other computational tools across six cancer types from seven cohorts of patients with superior sensitivity and specificity. It captures occult metastatic cells in a patient with a lymph node-negative breast cancer, who was declared to have no evidence of disease by conventional imaging methods but was later confirmed to have metastatic disease. Notably, EmitGCL identifies HSP90AA1 and HSP90AB1 as predictable biomarkers for future breast cancer metastasis, which we validate by in-vitro pharmacological inhibition of HSP90 that reduced breast cancer cell migration and further support across five independent cohorts of patients (n = 420). Furthermore, we demonstrate YY1 transcription factor as a key driver of breast cancer metastasis, which we corroborate with in-silico, CRISPR-based migration assays, and in vivo mouse lung colonization experiments, suggesting that YY1 is a potential therapeutic target for further investigation.

Predicting future metastases remains a major clinical challenge. Here, the authors develop EmitGCL, a deep-learning framework to predict metastasis and related biomarkers using cancer single-cell sequencing data, enabling and validating the discovery of occult metastases and breast cancer metastasis biomarkers.

Introduction

Advances in multi-modality therapy, such as surgery, radiation therapy, chemotherapy, targeted therapy, and immuno-therapy, have improved survival outcomes in cancer patients, yet metastatic disease remains a major challenge^1,2. Clinical studies have primarily focused on identifying predictive biomarkers for metastatic risk, such as (i) imaging-based markers, including radiomic features extracted from MRI and PET scans, (ii) HER2 enrichment³ which has been identified as a predictor of brain metastasis in breast cancer patients, and (iii) circulating tumor cells, which emerging studies suggest may serve as predictive markers for distant organ involvement⁴. However, most biomarkers were identified based on detectable metastatic cells^5–7, and they fail to capture cancer cells in metastatic sites that have already spread but remain undetectable due to their rarity⁸ (i.e., occult metastatic cells) and those with the potential to initiate metastasis at the primary site⁹ (i.e., metastatic precursor cells). As a result, existing biomarkers are often prone to overdiagnosis (i.e., false positives) or to missing actual metastases (i.e., false negatives), which can lead to unnecessary treatments or delayed diagnosis¹⁰. Computational studies identify predictive biomarkers through differentially expressed genes based on bulk RNA-seq data^11–14. Still, no common biomarkers across different cohort groups of patients have been identified, limiting their generalizability and effectiveness^15,16. These underscore the need for more sensitive and specific biomarkers to determine whether a cancer patient with no detectable metastasis is likely to develop it over time (i.e., future metastasis)⁶.

With advancements in single-cell RNA sequencing (scRNA-seq), researchers can now detect various cell types and states, including rare subpopulations, and characterize cellular heterogeneity of cancer at high resolution^17,18. However, identifying metastatic precursor cells with high metastatic potential within primary tumors and occult metastatic cells at metastatic sites remains challenging for existing computational methods and pathological examination techniques¹⁹. This limitation arises from (i) low sensitivity combined with high FPRs in identifying occult metastatic cells and (ii) the inability to trace the dynamic metastatic transition between precursor and occult metastatic cells. More broadly, case-versus-control classifications oversimplify diseases that exist on a spectrum, underscoring the need for models that capture continuous and dynamic transitions²⁰. Advancements in artificial intelligence (AI) offer a potential solution by enabling deep feature extraction and pattern recognition in single-cell omics data analysis^21–23.

Here, we present EmitGCL (Early Metastasis Identification and Therapeutic Discovery via Graph Contrastive Learning), a knowledge-aware graph contrastive learning framework that predicts future metastasis and identifies clinically actionable biomarkers and therapeutic targets from single-cell omics data. Across 15 scRNA-seq datasets spanning pancreatic, nasopharyngeal, papillary thyroid, head and neck, and breast cancers, EmitGCL achieves the highest true positive rate (TPR) for detecting occult metastatic cells, improving the second-best method by 12.44%. It accurately identifies occult metastatic cells in a lymph node-negative breast cancer patient that are subsequently confirmed as metastatic while maintaining a 0% false positive rate (FPR) in non-metastatic patients, enabling more precise metastatic risk stratification. Beyond metastatic cell identification, EmitGCL identifies HSP90AA1 and HSP90AB1 as biomarkers of future breast cancer metastasis, improving metastatic cell identification accuracy by 11–16% over differential expression analysis, and validates their functional relevance through pharmacological inhibition and independent clinical cohorts. Furthermore, EmitGCL identifies YY1 as an upstream therapeutic target through regulatory network analysis, and its functional role is validated using in silico transcription factor perturbation, CRISPR-mediated perturbation, in vitro migration assays, an in vivo mouse metastasis model, and multiple independent clinical cohorts. Together, these results demonstrate that EmitGCL provides a unified framework for early metastasis prediction, biomarker discovery, and therapeutic target identification, thereby facilitating precision oncology and metastasis prevention.

Results

Dataset characteristics and system overview

EmitGCL is a knowledge-aware contrastive learning approach to identify metastatic precursor cells and biomarkers. The project emphasizes AI modeling, bioinformatics analysis, in-vitro and in-vivo experimental validation, and clinical cohort data support to ensure the accuracy and applicability of its findings (Fig. 1a and Supplementary Fig. 1). To improve the signal-to-noise ratio in matched scRNA-seq data from primary and metastatic sites, we constructed a heterogeneous graph where nodes represent cells and genes and edges indicate gene presence in cells, enabling information propagation across shared gene–cell neighborhoods that amplifies consistent biological signals while suppressing stochastic noise (Fig. 1b). Based on the hypothesis that metastatic precursor cells are more similar to occult metastatic cells within the same clonotype than to other cells, we predict future metastasis more reliably than existing metastatic risk prediction methods by: (i) enhancing TPR, which indicates more accurately identifying occult metastatic cells from metastatic patients. This is achieved through a graph contrastive learning approach, which distinguishes subtle differences in the same tumor clonotype between primary and metastatic sites (Fig. 1c), (ii) reducing FPR, which indicates minimizing misclassification of non-metastatic patients as metastasis. This is accomplished by integrating prior knowledge from metastasis-related biological process databases (Fig. 1d) and (iii) identifying metastatic precursor cells by tracing occult metastatic cells through site-specific contrastive learning (Fig. 1e).

Fig. 1. Workflow of the EmitGCL framework.

a Various omics datasets were collected from public databases and literature. EmitGCL was trained using heterogeneous graphs with knowledge-aware contrastive learning to predict precursor metastatic cells and biomarkers. Multiple genes and TFs were identified as markers or therapeutic targets for future metastasis prediction, which were validated using experimental and clinical cohorts. Blue means primary site. Pink means metastatic sites. Circles represent cells, and squares represent genes. b scRNA-seq data from both were collected to construct a heterogeneous graph. c The contrastive learning process amplifies differences between primary and metastatic sites within the same clonotype. d The metastatic pathways were involved in model training by penalizing cells with low metastatic potential scores. e The system outputs include metastatic precursor cells and biomarkers in the primary tumor, as well as occult metastatic cells at metastatic sites. Source data are provided with this paper. Created in BioRender. Wu, W. (2026) https://BioRender.com/uwfr1hv.

We collected all publicly available matched scRNA-seq datasets from primary and metastatic sites, resulting in 33 paired datasets across breast, pancreas, nasopharyngeal, papillary thyroid, and head and neck cancers. These datasets were derived from seven published studies, involving 28 patients and a total of 516,093 cells. Of these, 215,443 cells are from primary sites, and the other 300,650 are from metastatic sites. To predict future metastasis, it is crucial to identify metastatic precursor cells, as they will serve as the metastatic niche for future metastasis development. However, a clinically applicable biomarker that accurately predicts precursor cells requires precise identification of occult metastatic cells, as they provide the key linkage between precursor cells and actual metastatic progression. Furthermore, identifying biomarkers of metastatic precursor cells provides clinically applicable detection methods for future metastasis assessment. To systematically evaluate the performance of occult metastatic cell identification, we applied our model, which integrates metastatic prior knowledge, to identify occult metastatic cells. Based on the model’s predictions, we then constructed a benchmarking framework to assess performance in terms of TPR and FPR. Additionally, to validate the clinical impact of our model’s inferred biomarkers in metastatic precursor cells, we performed in vitro functional validation using pharmacological inhibition of HSP90, which significantly reduced breast cancer cell migration. We then collected data from 420 breast cancer patients with metastasis across five independent cohorts and conducted survival analysis, further supporting the clinical relevance of the identified biomarkers (Fig. 1a)²⁴. To validate TF candidates that drive metastasis, we integrated CHIP-seq data and performed in-silico TF knockout experiments using CellOracle, CRISPR-based perturbation, migration assays, in vivo mouse lung colonization assay, and survival analysis across four independent clinical cohorts.

Evaluation of computational methods for identifying occult metastasis cells

Firstly, we assessed the performance of EmitGCL in identifying occult metastasis cells using matched scRNA-seq data from primary and metastatic sites across six cancer types (Fig. 2a). Among the 28 patients with 33 paired scRNA-seq datasets, metastasis was observed in five of the six cancer types (Supplementary Data 1). The exception was a case of lung cancer from a single patient with two lymph nodes, which did not exhibit metastatic progression. Pancreatic cancer was associated with liver metastasis, while the other four types (breast, nasopharyngeal, papillary thyroid, and head and neck) demonstrated lymph node metastasis (Fig. 2b, c). To validate our model, we compared EmitGCL against four published methods based on their performance reported in the study²⁵. MarsGT demonstrated the best performance for rare cell identification in that study. Seurat²⁶ serves as the baseline for cell clustering. GiniClust¹⁷ and CellSIUS²⁷ were the second-best tools for rare cell identification using clustering and classification methods, respectively, as evaluated by adjusted rand index (ARI) and F1-score.

Fig. 2. Data collection and computational evaluation of the model for occult metastatic cell prediction.

a Schematic depicting the study design. Created in BioRender. Wu, W. (2026) https://BioRender.com/uwfr1hv. Published paired scRNA-seq data from six cancer types were used. Blue circles represent primary sites, pink circles represent metastatic sites, and green circles represent cancer non-metastasis. b Pie chart showing data distribution. The outer circle represents metastasis versus non-metastasis, the middle segment shows metastatic sites, and the inner circle depicts the distribution of cancer types. c Bar plots summarizing the number of samples, datasets, and cells collected, categorized by primary organ and metastatic sites. d False positive evaluation of occult metastatic cell identification in lung cancer datasets without metastasis. The bar plots represent the number of identified metastatic precursor cells in primary tumors and associated metastatic cells in the lymph node. The line plots display the proportion of metastatic cells among all cells in the metastatic site. e True positive analysis of metastatic cell identification across four cancer types. Metrics include CNV, EMT activity, stemness, cell group proportion, and enrichment scores of metastasis-related pathways within metastatic precursor and occult metastatic cells. These evaluations are compared across different methods for each cancer type. f Overall performance score for metastatic cell identification in each cancer type. The y-axis indicates the proportion of correctly identified metastatic datasets relative to the total datasets for each cancer type. g Accuracy evaluation of metastatic cell identification across all datasets. Bar plots represent accuracy scores for each method. h False positive evaluation of metastatic cell identification across all datasets. Bar plots depict FPR for each method. Source data are provided with this paper.

To evaluate the FPR in identifying occult metastatic cells, we used two paired scRNA-seq datasets from primary lung cancer and lymph node sites that did not show metastasis. The results indicated that although all other tools misclassified non-occult metastatic cells as occult metastatic cells in these datasets, EmitGCL did not identify any occult metastatic cells, demonstrating its low FPR in identifying occult metastatic cells (Fig. 2d, Supplementary Data 2, and Source Data 1). Furthermore, to validate the TPR of occult metastasis identification, we evaluated EmitGCL alongside other tools across pancreas, nasopharyngeal, papillary thyroid, and head and neck cancers that have metastasized. We visualized several key biological features in the identified occult metastatic cells, including copy number variation (CNV)^28,29, epithelial-to-mesenchymal transition (EMT) activity scores³⁰, tumor stemness scores³⁰, and metastasis-related pathway enrichment scores³¹. EmitGCL consistently outperformed other tools, with the identified occult metastatic cells showing higher performance across these metrics (Fig. 2e, Supplementary Fig. 2, Supplementary Data 3–8, and Source Data 2–6). Next, we calculated the rate of occult metastatic cell detection across all datasets for the above four cancer types, which demonstrated that EmitGCL has a higher TPR in identifying occult metastatic cells compared to other methods (Fig. 2f). To provide an overall summary of performance, we compared the TPR and FPR across all datasets (“Methods”), showing that it effectively reduces FPR while maintaining high TPR (Fig. 2g, h, Supplementary Data 9, and Source Data 7).

To further assess EmitGCL’s robustness, we ran the model five times for each cancer type and evaluated consistency using the ARI, where a higher ARI indicates greater model robustness. It achieved ARI scores higher than 0.95 across all runs, confirming its reproducibility (Supplementary Fig. 3a). We then performed 20 random initializations of the model parameters, and the resulting ARI values demonstrated the stability of EmitGCL (Supplementary Fig. 3b). To further assess robustness for rare cell populations, we computed F1 scores across the same 20 runs, which also remained stable (Supplementary Fig. 3c). In addition, to evaluate gene-level robustness, we calculated the overlap of the top 50 genes identified in each run; the overlaps ranged from 72 to 94% across datasets (Supplementary Fig. 3d). EmitGCL was validated on held-out test cells using ARI and NMI, which remained high, confirming good generalization without overfitting (Supplementary Fig. 3e). EmitGCL’s four loss terms remained balanced during training, and a grid search across 900 weight combinations showed the model’s performance (ARI/NMI) was robust and insensitive to weight settings (Supplementary Fig. 3f–h). Additionally, we validated that EmitGCL-identified biomarker genes, including known metastasis-related genes like B2M³² and MALAT1³³, were consistently detected across primary and metastatic sites, aligning with previous findings (Supplementary Fig. 4 and Supplementary Data 10).

Validation of occult metastatic cell prediction based on clinical trace data

In the benchmark section (Fig. 1e and Supplementary Fig. 2), we demonstrated that EmitGCL can fully identify patients with metastasis across four cancer types. To further validate our model’s ability to detect metastasis, we analyzed a representative case: paired scRNA-seq data from a breast cancer patient with a primary tumor and two lymph nodes (LN-1 and LN-2), which were initially deemed negative but later confirmed as metastatic, despite being undetectable by conventional pathological evaluation³⁴. EmitGCL identified cell cluster 2 in LN-1 and cell cluster 4 in LN-2 as metastatic precursor and occult metastatic cells (Fig. 3a). The corresponding cell counts in the primary and metastatic sites were 995 and 158 in LN-1, and 1109 and 23 in LN-2, respectively (Fig. 3b). The proportion of occult metastatic cells in the metastatic sites was less than 3%, highlighting the rarity of these cells’ population (Fig. 3c). Further analysis revealed that these clusters exhibited high CNV scores and cancer cell stemness scores, distinguishing them from other cell clusters in LN-1 and LN-2 (Fig. 3d, e). Additionally, the EMT scores in the identified occult metastatic cells were significantly higher than those in the metastatic precursor cells, providing further evidence that cell cluster 2 in LN-1 and cell cluster 4 in LN-2 represent true occult metastatic populations (Fig. 3f and Supplementary Data 11, Source Data 8).

Fig. 3. EmitGCL captures the occult metastatic cells in breast cancer metastasis patients with negative status lymph nodes.

a UMAP visualizes cell clusters predicted by EmitGCL in breast cancer with two negative lymph nodes (LN-1 and LN-2). The cluster with the red circle represents the metastatic precursor cell in the primary tumor and the occult metastatic cells in the lymph node identified by EmitGCL. b Bar plots show the cell number of the identified precursor (blue) and occult metastatic cells (pink) in LN-1 and LN-2. c Bar plots show the proportion of occult metastatic cells among all cells in the lymph node. d UMAP showcases the CNV score across all cells in LN-1 and LN-2. e Box plots showcase the tumor cell stemness score across all cell clusters identified by EmitGCL in LN-1 and LN-2. f Box plot showcases the EMT enrichment score in primary (P) and metastatic sites (M) across all cell clusters identified by EmitGCL in LN-1 and LN-2. Each box showcases the minimum, first quartile, median, third quartile, and maximum EMT enrichment scores on different sites (LN-1: P: N = 995, M: N = 158, LN-2: P: N = 1109, M: N = 23). The p-value is calculated by the Mann–Whitney U test with one-sided. g The pseudotime of predicted precursor in primary tumor and associated metastatic cells in LN-1 and LN-2. Blue means the cell is located in the primary, and pink means the cell is located in the metastatic sites. h Box plot showcases the metastasis-related pathway enrichment score in primary (P) and metastatic sites (M) across all cell clusters identified by EmitGCL in LN-1 and LN-2. Each box showcases the minimum, first quartile, median, third quartile, and maximum metastasis-related pathway enrichment scores on different sites (LN-1: P: N = 995, M: N = 158; LN-2: P: N = 1109, M: N = 23). N is the cell number. Statistical significance was assessed using a one-sided Mann–Whitney U test. For the Hippo pathway, LN-1 showed U = 70,981 and P = 0.025, whereas LN-2 showed U = 8723 and P = 0.0146. N denotes the number of cells. Source data are provided with this paper.

Given that metastatic cells could exhibit characteristics of later stages in cancer cell evolutionary progression²⁸, we visualized the cellular trajectories and corresponding pseudotime for cluster 2 in LN-1 and cluster 4 in LN-2, respectively (Fig. 3g and Supplementary Fig. 5a). This visualization further supports the identification of these clusters as true positives. Moreover, metastasis-related pathway enrichment scores were significantly higher in metastatic sites than in the primary site, consistent with the established knowledge that metastatic potential increases in cancer cells at metastatic sites. This may be driven in part by dysregulation of the Hippo signaling pathway, where increased YAP/TAZ activity promotes cancer cell survival, proliferation, and enhanced metastatic capacity^35–37 (Fig. 3h, Supplementary Fig. 5b, and Supplementary Data 12, Source Data 9). These findings provide strong evidence that EmitGCL can accurately identify occult metastatic cells that may be missed by conventional clinical methods, as well as metastatic precursor cells in primary tumors.

Validation of biomarkers in precursor cells based on in-vitro experiment and clinical cohort data

Previous validation demonstrated the superior performance of EmitGCL in identifying occult metastatic cells, providing a reliable foundation for metastatic precursor cell prediction. To identify biomarkers in metastatic precursor cells, we trained EmitGCL on data from 18 breast cancer patients. During model training, attention values were analyzed to identify biomarker genes specific to metastatic precursor and occult metastatic cells (“Methods”) (Fig. 4a and Supplementary Data 13, Source Data 10). Many of the genes identified in both metastatic precursor and occult metastatic cells, such as ACTB³⁸, have already been validated as highly associated with breast cancer metastasis. The unique biomarkers in metastatic precursor cells identified from primary tumors were then used to build a regression model aimed at predicting future metastasis (“Methods”).

Fig. 4. EmitGCL efficiently identifies breast cancer metastatic precursor gene biomarkers, demonstrating the potential for predicting the timing of occult metastasis in clinical settings.

a Dot plot showing the attention scores of gene biomarkers in the metastatic precursor (blue) cells in the primary tumor and occult metastatic cells in the lymph node. Colors represent different sites, and dot size indicates attention score. b Bar plots showing the attention scores of gene biomarkers identified exclusively in metastatic precursor cells. c Bar plots showing frequency of these genes across five metastatic breast cancer cohorts. d EO771 cells were subjected to a scratch-wound migration assay in the presence or absence of the HSP90 inhibitor STA-9090 (10 or 100 nM) and imaged after 30 h. Scale bars are 1000 μm, which are shown in the lower right corner of each panel. e Wound closure was quantified using FIJI and normalized to cell viability measured under identical conditions to control for cytotoxic effects. Statistical analysis was performed using one-way ANOVA with Tukey’s post-hoc test (n = 3 biological replicates per group: vehicle, 10 nM, and 100 nM STA-9090). Data are presented as mean ±  s.e.m. f Metastasis-free survival curve stratified by predicted future metastasis results using attention core of HSP90AA1 and HSP90AB1. Groups were compared using a one-sided log-rank test. g Bar plot showing the accuracy of predicting the development of metastasis and those without future metastasis using a train/test dataset split (9:1), based on biomarkers identified by tracing clinically evident metastatic genes. h Box plot showing the accuracy of predicting high versus low metastatic potential using a leave-one-dataset-out approach. Each box showcases the minimum, first quartile, median, third quartile, and maximum accuracy (Caldas: N = 52, Chin: N = 95; Desmedt: N = 75; Miller: N = 83, and TCGA: N = 115). N is the sample number. Groups were compared using a one-sided Wilcoxon test. i Metastasis-free survival curve stratified by predicted future metastasis results using the attention core of ACTB. Groups were compared using a one-sided log-rank test. j Bar plot comparing the performance for predicting biomarkers identified by EmitGCL, Zhou’s TCGA-based method, a cell clustering method based on DEGs (DEG1), and a cell classification method based on DEGs (DEG2). Source data are provided with this paper.

Among the identified biomarkers, HSP90AA1, HSP90AB1, and H3F3B were consistently observed across five independent cohorts and exhibited high attention scores (Fig. 4b). Notably, these biomarkers had contrasting fold-change values, rendering them undetectable through traditional differential expression analysis (Supplementary Fig. 6a, b). To assess the accuracy and robustness of these biomarkers, we applied different train/test dataset splits and analyzed combinations of the three biomarkers. EmitGCL showed high accuracy across all splits, particularly in the 9:1 configuration, indicating robust performance. The combination of HSP90AA1 and HSP90AB1 delivered the best results and was selected for further analysis (Fig. 4c, Supplementary Fig. 6c, and Supplementary Data 14, Source Data 11). To functionally validate the metastasis-associated role of the HSP90 axis predicted by EmitGCL, EO771 breast cancer cells were subjected to a scratch-wound migration assay in the presence or absence of the HSP90 inhibitor STA-9090 (ganetespib) (Fig. 4d). Treatment with 10 nM STA-9090 for 30 h had minimal impact on cell migration, whereas 100 nM treatment significantly reduced wound closure (Fig. 4e). To control for potential cytotoxic effects, parallel viability assays were performed under identical treatment conditions. Treatment with 100 nM STA-9090 resulted in only a modest reduction in cell viability (~10%). Wound closure measurements were therefore normalized to viability to isolate migration-specific effects. After normalization, HSP90 inhibition remained associated with a significant reduction in migratory capacity, indicating that the observed effect is not attributable to decreased cell viability.

To further assess the clinical relevance of the HSP90AA1–HSP90AB1 axis identified by EmitGCL and supported by functional experiments in mouse models, we next evaluated whether these biomarkers were associated with patient outcomes. A survival analysis based on future metastasis predictions using HSP90AA1 and HSP90AB1 revealed that patients who were predicted to have future metastasis had significantly shorter overall survival compared to those without future metastasis (Fig. 4f and Supplementary Data 15, Source Data 12).

We hypothesize that metastatic precursor cells are better traced through occult metastatic cells rather than clinically evident metastatic cells. To validate this, we identified genes in metastatic precursor cells by tracing them through clinically evident metastatic cells (Supplementary Fig. 6d). The single gene ACTB yielded the best results and was selected for further analysis and comparison (Fig. 4g). The validation using a leave-one-dataset-out approach across five cohorts also confirmed the consistent predictive accuracy of EmitGCL when trained on occult and clinically evident metastatic biomarkers (Fig. 4h, and Supplementary Data 16, Source Data 13). Simultaneously, survival analysis using ACTB showed no significant difference in overall survival between patients predicted to develop metastasis and those without future metastasis (Fig. 4i and Supplementary Data 17, Source Data 14), confirming that EmitGCL based on clinically evident metastatic cells cannot reliably predict biomarkers of future metastasis.

To further validate the advantage of our method over the widely used DEG-based approaches, we examined DEGs between short-term and long-term metastases in each sample and found that no common DEGs were present across the five cohorts, and the large number of DEGs identified were unsuitable for clinical application due to their lack of specificity (Supplementary Fig. 6e).

To provide an overall performance evaluation, we compared EmitGCL to other methods relying on DEGs. EmitGCL outperformed approaches based on clinically evident metastatic cells, as well as Zhou’s TCGA-based methods⁶, clustering-based (DEG1), and classification-based (DEG2) methods in predicting future metastasis (Fig. 4j). In conclusion, this case study highlights the effectiveness of EmitGCL in identifying biomarkers of metastatic precursor cells and predicting future metastasis. This approach ultimately informs treatment strategies and improves patient prognosis.

Multi-level validation of candidate transcriptional regulators associated with metastatic potential

Based on the genes identified in (Fig. 4a), the gene regulatory networks in both sites were inferred using the ChEA3 database³⁹. The importance of TFs was ranked across multiple metrics, and the high-ranked TFs will undergo simulation and experimental validation (Supplementary Fig. 7a). The resulting regulatory networks in primary and metastatic sites are displayed in Fig. 5a. We evaluated all regulators and regulons based on TF frequency, TF expression, and regulon attention scores, identifying 12 regulons with high scores across all indices (Fig. 5b and Supplementary Data 18, Source Data 15). Among them, JUN⁴⁰ and FOS⁴¹ had the highest scores and have been reported to be associated with metastasis in previous studies. To further validate TFs that have not yet been confirmed in existing publications, we selected YY1, which demonstrated high TF closeness within the metastatic regulatory networks and is also a component of the polycomb repressive complex (PRC). PRC is a key epigenetic regulatory complex involved in chromatin remodeling and transcriptional regulation and has been implicated in oncogenic transcriptional reprogramming⁴². Another PRC component, EZH2, has been shown to drive breast cancer metastasis via integrin β1-FAK activation⁴³. However, the role of YY1 in this process remains poorly understood.

Fig. 5. EmitGCL experimentally and clinically validates critical TFs for inducing breast cancer metastasis.

a TF regulatory networks of precursor and occult metastatic cells across breast cancer patients; colours indicate precursor, occult metastatic and common relationships. b Heatmap of TF frequency, expression, regulon activity score, degree, and closeness centrality for the network in (a). c Observed and extrapolated future states (arrows) after YY1 knockout in four epithelial subpopulations; colours denote cell clusters. d Wound-healing assay of YY1KO versus control cells at 24 h. Scale bars are shown in the lower right corner of each panel. Scale bars are 300 μm. e Wound-healing area, YY1KO versus control (n = 6 independent biological replicates per group, each derived from a separate CRISPR knockout reaction, not technical/well replicates; two-way ANOVA with multiple-comparisons correction; 0 h P = 0.8162 (ns), 24 h P = 0.0001). f Migration capacity, YY1KO versus control (n = 6 per group); wild-type versus YY1KO adjusted P = 0.0018. g Relative YY1 mRNA by qPCR, Control versus Yy1KO (n = 3 independent biological replicates; two-sided unpaired t-test; exact P = 0.03). h Representative transwell migration images, parental EO771 versus YY1-knockout cells. Scale bars are 100 μm, which are shown in the lower right corner of each panel. i Quantification of invading cells, parental EO771 (control, n = 5) versus YY1-knockout (sg-Yy1, n = 6); one-way ANOVA F(3,19) = 26.85, P < 0.0001, R² = 0.81, with Dunnett’s test versus control; control versus sg-Yy1 mean difference 188.8 (95% CI 127.3–250.3), adjusted P = 7.1 × 10⁻⁷. j Control and Yy1-overexpressing EO771 cells were injected into C57BL/6 mice by tail-vein (n = 6 per group); lung metastasis was assessed at day 15 by in vivo bioluminescence. k Total photon flux quantifying metastatic burden (n = 6 mice per group). Kaplan–Meier survival by median YY1 expression, each with a box plot of YY1 expression in high- versus low-expression groups; hazard ratio (95% CI) and one-sided log-rank P: l GSE2990 relapse-free (n = 125), 1.59 (1.00–2.52), 0.049; m GSE6532 relapse-free (n = 125), 1.62 (1.02–2.57), 0.042; n GSE6532 relapse-free (n = 54), n is the patient number; 1.87 (1.00–3.51), 0.050; o GSE6532 overall (n = 155), 2.08 (1.08–4.03), 0.029. Data are mean ± s.d. (e, g) or mean ± s.e.m. (f, i, k); all replicates are biological and independent (not technical replicates). Tests: two-way ANOVA with multiple-comparisons correction (e); two-sided unpaired t-test (g); one-way ANOVA with Dunnett’s test (f, i); two-sided Welch’s t-test (k); one-sided Wilcoxon test for box plots (l–o). Box plots: centre, median; bounds, first–third quartiles; whiskers, 1.5× IQR; outliers shown individually. Source data are provided with this paper.

We initially conducted in-silico perturbation using CellOracle⁴⁴ for each breast cancer patient. The cell clusters identified by EmitGCL for each patient were visualized using UMAP, with cluster annotations determined based on the marker genes corresponding to each cell type (Supplementary Figs. 7b and 8a). Based on the identified cell clusters, epithelial and cancer cells were extracted as inputs for simulated knockout validation using CellOracle. The results revealed that cells with high metastatic potential showed a predicted shift toward a less aggressive epithelial-like state, consistent with a potential role of YY1 in regulating metastasis-associated transcriptional programs (Fig. 5c and Supplementary Fig. 8b).

To exclude a viability confounder, we measured cell viability in the exact time window used for our in vitro wound-healing and Transwell assays (48 h) using a CCK-8 assay. YY1 knockout did not reduce viability compared with control in EO771 cells (n = 3; mean ± SEM A450:0.521 ± 0.005 vs. 0.511 ± 0.007; two-sided unpaired t-test, p = 0.2871). These data indicated that the genetic deletion of YY1 has no impact on the survival of breast cancer cell lines, thus the YY1-mediated migration is unlikely to be attributable to cell death in our experiments (Supplementary Fig. 9a and Source Data 18). To further validate the role of key TFs in the occult metastasis process, we used CRISPR-mediated knockout of FOXP1 and JUND in EO771 cells and demonstrated a significant deficiency in cell migratory capacity by transwell cell migration assay (Supplementary Fig. 9b–d and Supplementary Data 19, Source Data 16). This finding aligns with previous studies implicating these TFs in cancer progression and metastatic processes. Subsequently, we focused on the identified TF, YY1, and observed that YY1^KO cells exhibit reduced wound closure at 24 h compared to the control group (Fig. 5d, Supplementary Fig. 10 and Supplementary Data 19). Quantification of wound healing areas and wound closure speed revealed a significant reduction in YY1^KO cells compared to control groups (**p < 0.01, ***p < 0.001) (Figs. 5e, f, and Source Data 17). YY1 mRNA expression measured by qPCR indicated a significant reduction in YY1 expression in YY1^KO compared to control (*p < 0.05) (Fig. 5g). To more accurately investigate the ability of tumor cells to invade across the basement membrane, rather than merely assessing lateral migratory or proliferative capacity as in wound healing assays, we performed a transwell cell migration assay and revealed that YY1^KO cells had a significant reduction in cell migratory capacity compared to the control group (****p < 0.0001) (Fig. 5h, i).

To determine whether Yy1 also influences metastatic potential in vivo, we generated luciferase-expressing EO771 murine breast cancer cells with stable overexpression of Yy1 or an empty vector control (Supplementary Fig. 9e). Cells (1 × 10⁵) were injected into 6–8-week-old female C57BL/6 mice via lateral tail-vein injection (n = 6 per group). Bioluminescent imaging at 15 days post-injection indicated greater bioluminescence in the thoracic region for Yy1-expressing cells (Fig. 5j). Quantification of photon flux between the groups confirmed a significantly higher metastatic burden in the lungs of mice with Yy1-expressing cells compared to mice injected with control EO771 cells (Fig. 5k). Together, these results demonstrate that Yy1 enhances metastatic colonization in vivo, providing functional evidence supporting its role as a driver of breast cancer metastasis.

To further validate the role of YY1 in the survival outcomes of breast cancer patients, four independent cohort studies with bulk RNA sequencing data were analyzed. Patients in each cohort were stratified into high and low YY1 expression groups based on the median expression level. Relapse-free survival was assessed in cohorts 1–3, while overall survival was evaluated in cohort 4, using Kaplan–Meier curves for all analyses. Survival differences were compared using the log-rank test. In cohorts 1–3, patients with YY1 expression levels higher than the median exhibited significantly shorter relapse-free survival, with hazard ratios of 1.59 [95% confidence interval (CI), 1.00–2.52] (p = 0.049), 1.62 [95% CI, 1.02–2.57] (p = 0.042), and 1.87 [95% CI, 1.00–3.51] (p = 0.050) (Fig. 5l–n), respectively. Similarly, patients with higher YY1 expression in cohort 4 demonstrated significantly shorter overall survival, with a hazard ratio of 2.08 [95% CI, 1.08–4.03] (p = 0.029) (Fig. 5o).

EmitGCL reveals a more immunosuppressive environment in lymph node sites compared to primary sites in breast cancer

To understand the differences between primary and metastatic sites during the metastatic process, we performed pathway enrichment analysis on gene signatures from both locations (Fig. 6a). The results revealed that immune-related pathways, such as antigen processing and presentation, were more significantly enriched in the primary site, suggesting a more active immune environment compared to the metastatic site. To explore the underlying reasons for this observation, we used CellChat to analyze cell-cell communication between early metastatic cell groups and cytotoxic T cells, as well as between early metastatic cell groups and antigen-presenting cells. Our analysis showed that the MIF signaling pathway, known for its immunosuppressive effects, had a higher communication probability between early metastatic cell groups and cytotoxic T cells compared to precursor cell groups and cytotoxic T cells (Fig. 6b). Further, we calculated the T cell exhaustion and effective activity scores at both primary and lymph node sites. The results revealed that T cells exhibited higher exhaustion scores in lymph nodes, indicating a more immunosuppressive environment, while showing higher effective activity in primary sites (Fig. 6c). This was corroborated by the bar plot of T cell clonotypes, where lymph nodes demonstrated more clonotype expansion than primary sites, highlighting the progression of immune suppression in metastatic environments (Fig. 6d–f). Additionally, pathway enrichment analysis based on differentially expressed genes (DEGs) between precursor and metastatic cell groups further supported the presence of enhanced immunosuppressive mechanisms in the lymph node sites (Fig. 6g).

Fig. 6. EmitGCL reveals a more immunosuppressive environment in lymph node sites compared to primary sites in breast cancer.

a Pathway enrichment analysis across precursor (blue) and early metastatic cell groups (pink). Colors represent different sites, and dot size indicates the ratio of enriched genes. p-values were calculated using a one-sided hypergeometric test, with FDR-adjusted p-values for multiple testing correction. b MIF signaling pathway analysis of cell-cell communication between precursor and early metastatic cell groups and cytotoxic/antigen-presenting cells. Colors represent communication probability. c Enrichment scores of effective and exhausted gene signatures across different sites in cytotoxic T cells (primary: $\mathit{n}$ = 176, metastasis: $\mathit{n}$ = 895, $\mathit{n}$ represents cell number). Box plots show the median (centre line), the first and third quartiles (box bounds), and whiskers extending to 1.5× the interquartile range, with outliers plotted individually. Statistical significance was assessed using a one-sided Mann–Whitney U test. The exhausted activity score was significantly higher in metastatic sites than in primary sites (W = 10,789,674, P = 0.0094), whereas the effector activity score was significantly higher in primary sites than in metastatic sites (W = 42,244,096, P < 2.22 × 10⁻¹⁶). d Bar plot showing the number of T clonotypes in primary and metastatic sites. e Cell location across all samples based on scTCR-seq data. UMAP plot showing the distribution of cells between the primary tumor (green) and lymph node metastasis sites (blue). f Clone expansion across primary and lymph node sites. UMAP plot highlighting T cell clonal expansion, with color intensity (red) indicating higher clonal expansion, showing greater expansion in lymph node sites compared to primary sites. g Immune-related pathway enrichment based on DEGs between precursor and metastatic cell groups. Dot size indicates the count of enriched genes. q-values were calculated using a one-sided hypergeometric test with multiple testing correction. Source data are provided with this paper.

Discussion

EmitGCL provides a useful approach for predicting future metastasis based on scRNA-seq, particularly in complex biological settings where traditional computational methods and clinical diagnostic technologies often fall short. By integrating prior knowledge of metastasis, EmitGCL achieves high TPR while maintaining a low FPR in occult metastatic cell identification. This accuracy provides the necessary foundation for reliable metastatic precursor cell inference. Leveraging graph contrastive learning, EmitGCL enables the tracing of dynamic metastatic transitions between metastatic precursor cells and occult metastatic cells.

In this study, which involved 28 patients and 516,093 cells across six cancer types, we evaluated the performance of EmitGCL in identifying occult metastatic cells through computational validation. While direct targeted attempts for the HSP90 complex have not been successful clinically, the HSP90 complex is highly critical to the integrity and function of many oncogenes by maintaining their protein folding and stability. These genes are primary markers of metastatic precursor cells. Their expression aligns with the understanding that metastasis is a highly stressful process for cancer cells, where the cellular response to stress and the maintenance of protein integrity are likely key to successful dissemination. To functionally validate this prediction, we performed pharmacological inhibition experiments using the HSP90 inhibitor STA-9090, which significantly reduced breast cancer cell migration in vitro even after controlling potential cytotoxic effects, supporting a functional role for HSP90AA1 and HSP90AB1 in regulating metastatic phenotypes. However, these genes could prove to be good biomarkers for the likelihood of metastasis and delineate whether primary tumors need additional treatment to prevent future metastasis from occurring. Additionally, EmitGCL uncovered the key transcription factor YY1, which plays a role in driving breast cancer metastasis. To functionally validate this prediction, we performed both in vitro and in vivo experiments. Loss of YY1 reduced tumor cell migratory capacity in vitro, whereas stable overexpression of Yy1 significantly increased lung metastatic colonization in a tail-vein mouse model. These complementary loss- and gain-of-function results provide functional evidence supporting a causal role for YY1 in promoting metastasis. YY1 has been shown to play various roles in tumors, including regulating the expression of oncogenes and tumor suppressor genes, metabolic reprogramming, and promoting the cancer stem cell phenotype. YY1 has also been shown to regulate the tumor microenvironment by promoting angiogenesis and suppressing immune-mediated killing, among others. These functions accomplished by YY1 would be highly beneficial and likely necessary for metastasis to be successful. YY1 was likely detected by EmitGCL for these reasons and will provide promise for enhancing therapeutic strategies and providing insights for future clinical trial design.

Despite these promising results, some limitations of EmitGCL remain. First, a major limitation is the lack of prospective clinical trial validation for the biomarkers identified by EmitGCL. Without rigorous clinical testing, the translational potential of these findings remains speculative. Future studies will focus on validating these biomarkers in clinical trials to ensure their robustness and applicability in clinical settings. Such validation is critical for translating EmitGCL’s insights into actionable strategies in precision medicine, ultimately improving outcomes for patients at high risk of metastasis. Second, the current predictive model was developed based on data from seven cohorts comprising 28 patients, which exhibited considerable heterogeneity in clinical behavior. Future large-scale studies with a greater number of patients are needed to validate these findings and improve model generalizability. Third, in future work, adapting EmitGCL to spatially resolved transcriptomics represents a critical next step to further enhance biological credibility and clinical relevance. While current ST platforms face challenges, such as limited resolution and a lack of 3D context, our group has initiated a follow-up effort to integrate histopathology (H&E) with molecular references, inspired by spatial foundation models, such as OmiCLIP⁴⁵. In addition, while our framework is tailored to model malignant cell dissemination, evaluating model specificity against healthy tissue references could provide further insights. Resources, such as the Human Cell Atlas⁴⁶, which span diverse cell types ranging from proliferative to quiescent states, may serve as a valuable benchmark to test false-positive predictions in future work. We have highlighted this point as a potential direction for expanding validation efforts.

In conclusion, EmitGCL offers a prior knowledge-aware graph contrastive learning model capable of predicting future metastasis and this biomarkers. It also uncovers key TFs that can be regarded as therapeutic targets in clinical applications. EmitGCL provides valuable insights that could inform potential therapeutic strategies and clinical trial designs, ultimately holding promise for improving patient outcomes.

Methods

Ethics statement

This research complies with all relevant ethical regulations. All animal experiments were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of Indiana University Bloomington (protocol no. 23-025) and were conducted in accordance with institutional and federal guidelines for the care and use of laboratory animals. This study analyzed only previously published, publicly available human datasets; no new human samples were collected, and no additional ethical approval was required for their use.

Cell culture

EO771 cells (ATCC, Cat # CRL-3461) were maintained in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 2 mM glutamine, 100 U ml⁻¹ penicillin, 100 µg ml⁻¹ streptomycin and 10% fetal calf serum (FCS). Cells were authenticated by STR profiling and routinely checked for mycoplasma contamination.

Cell viability assay

To exclude a viability confounder, we measured cell viability in the exact time window used for our wound-healing and Transwell assays (48 h) using a CCK-8 assay.

Western blot

Yy1 overexpression in EO771 cells was confirmed by Western blot. Cells were lysed in RIPA buffer, and equal amounts of protein were separated by SDS-PAGE and transferred to a membrane. Membranes were probed with an anti-Yy1 antibody (Cell Signaling Technology, #46395) and an anti-β-actin antibody as a loading control, followed by an HRP-conjugated secondary antibody and chemiluminescent detection. The anti-Yy1 antibody has been validated by the manufacturer (Cell Signaling Technology) and in published studies.

CRISPR knockout

The CRISPR/Cas9 system was used to knock out the gene encoding transcription factor Yy1, Foxp1, and Jund in the EO771 cell line via electroporation. Two predesigned sgRNAs are selected to target each gene: Yy1 (5′-GCCCACCACCGTGGTCTCGA-3′ and 5′-ACCCTCTACATCGCCACGGA-3′), Foxp1 (5′-CTTCGTGACACTCGGTCCAA-3′, 5′-TAGTAAGTGGTTGCCACCGC-3′), and Jund (5′-TACGCAGTTCCTCTACCCGA-3′, 5′-GATCATCCAGTCCAACGGGC-3′), Cas9 Nuclease and electroporation enhancer were acquired from IDT. Electroporation was performed following the protocol provided by IDT CRISPR genome editing (www.idtdna.com) using the Lonza 4D Nucleofector System.

RNA extraction, reverse transcription, and quantitative PCR

Total RNA was extracted with the Zymo Research Direct-zol RNA Miniprep Plus kit (R2072). Reverse transcription was performed with LunaScript RT SuperMix (New England Biolabs, E3010), and quantitative PCR was carried out with Luna Universal qPCR Master Mix (New England Biolabs, M3003). PrimeTime predesigned qPCR assays were obtained from IDT for Yy1 (Mm.PT.58.7943099) and the reference gene Actb (Mm.PT.39a.22214843.g). Relative YY1 mRNA expression was normalized to Actb and calculated using the 2^−ΔΔCt method. Data are from n = 3 independent biological replicates.

Transwell migration assay

EO771 cells were subjected to CRISPR-based gene editing to generate knockout groups. Following transfection, cells were cultured for 48 h. Each group was then diluted to a concentration of 20,000 cells in 200 µl 10% fetal bovine serum (FBS)-containing medium and seeded into the upper chamber of transwell inserts. The lower chambers were filled with 600 µL of 10% FBS-containing medium. After incubation for approximately 36 h, inserts were collected, and non-invading cells on the upper surface of the membrane were gently removed using a cotton swab. The invaded cells were fixed in 70% ethanol for 1 min, followed by staining with 4% Trypan Blue for 10 min. The inserts were then washed twice with PBS. Images of invaded cells were captured using a light microscope for further analysis.

Wound healing migration assay

EO771 cells were subjected to CRISPR-based gene editing to generate knockout groups. Following transfection, cells were cultured for 48 h. Cells were grown to confluency in 6-well plates, followed by making the scratch, a vertical wound in the well, using a sterile pipette tip. Images were collected immediately after making the wound as the baseline time point. Experiments with CRISPR-based knockout were then grown for 24 h and imaged for the migration of cells into the wound. For experiments with Hsp90 inhibition, STA-9090 (Selleckchem; Cat # S1159) was added after baseline images were taken. Images were analyzed with FIJI to calculate the area of the wound that was closed during the 24 h incubation period.

RT-qPCR

Total RNA was extracted from control and Yy1-knockout EO771 cells using the Direct-zol RNA Miniprep Plus kit (Zymo Research, R2072) and reverse-transcribed with LunaScript RT SuperMix (New England Biolabs, E3010). Quantitative PCR was performed with Luna Universal qPCR Master Mix (New England Biolabs, M3003) using PrimeTime qPCR assays (Integrated DNA Technologies) for Yy1 (Mm.PT.58.7943099) and Actb (Mm.PT.39a.22214843.g) as the reference gene. Relative Yy1 expression was calculated using the 2^(−ΔΔCt) method (n = 3 independent biological replicates).

In vivo bioluminescence-based lung metastasis assay

EO771 female murine breast cancer cells were engineered to stably express firefly luciferase and either an empty vector (control) or a Yy1 overexpression construct. These constructs were used to generate a lentivirus that was used to infect EO771 cells that then underwent selection to arrive at the final engineered cell line with control or Yy1 expression. Constructs were designed and purchased from VectorBuilder (control: VB900172-2880vje; Yy1: VB900173-9278hhd). Prior to injection, cells were harvested during the logarithmic growth phase, washed with phosphate-buffered saline (PBS), and resuspended at the appropriate concentration in sterile PBS. All animal experiments were reviewed and approved by the Indiana University Bloomington Institutional Animal Care and Use Committee (IACUC) and were conducted in accordance with institutional and federal guidelines for the care and use of laboratory animals. Six- to 8-week-old female C57BL/6 (C57BL/6NHsd from Inotiv) mice were used for all experiments. Mice were housed in 12 h light/dark cycles with temperatures ranging from 18 to 26 °C at 40–60% humidity. The maximal tumour burden permitted by the IACUC was not exceeded in any experiment; because this tail-vein lung-colonization model does not produce a measurable subcutaneous tumour, animals were monitored using humane endpoints (body-weight loss and body-condition scoring), and no animal reached these limits. A total of 1 × 10⁵ viable cells in 100 µL sterile PBS were injected into the lateral tail vein of each mouse (n = 6 per group). To assess metastatic colonization, in vivo bioluminescence imaging was performed 15 days post-injection using an IVIS Lumina X5. Mice were anesthetized with isoflurane and administered D-luciferin (150 mg/kg, intraperitoneally). Photon flux (photons/sec) within the thoracic region was quantified using Living Image software version 4.8.4 as a measure of lung metastatic burden. Statistical comparisons between groups were performed using Welch’s t-test. Data are presented as mean ± SEM.

Ethics statement for tumor burden

Animal experiments were approved by the Bloomington Institutional Animal Care and Use Committee (IACUC). The maximal external tumor size permitted by the IACUC is 2000 mm³. For the experimental lung metastasis model (Fig. 5j), tumour burden could not be directly measured because metastatic nodules formed within the lungs following tail-vein injection. Tumour progression was therefore monitored by bioluminescence imaging, and animals were assessed according to predefined humane endpoints based on clinical signs of distress under the approved IACUC protocol. The maximal tumour burden permitted by the institutional guidelines was not exceeded.

Data description

We included 33 publicly available matched scRNA-seq datasets from primary and metastatic sites across 28 patients with six cancer types in the paper. All cells were used for model training. The paired lung and two lymph node scRNA-seq datasets from a lung cancer patient without metastasis were used for false positive evaluation in the benchmark section. The paired pancreatic and liver scRNA-seq data from pancreatic cancer patients with liver metastasis, along with data from four other cancer types (breast, nasopharyngeal, papillary thyroid, and head and neck), which exhibited lymph node metastasis, were used for sensitivity testing in the benchmark section. The paired breast and lymph node scRNA-seq datasets from 13 breast cancer patients with metastasis were used for case analysis. Additionally, 420 bulk breast cancer RNA-seq datasets from UCSC were used to validate the biomarkers identified by EmitGCL. All data are publicly available (Supplementary Data 1).

Model’s inputs

EmitGCL initiates by inputting two count matrices $X^P=\left\{x_{ij}^P\mid i=\mathrm{1,\; 2},\dots ,I;j=\mathrm{1,\; 2},\dots ,J_P\right\}$ and ${\mathrm{X}}^{\mathrm{M}}=\left\{x_{ij}^M\mid i=\mathrm{1,\; 2},\dots ,I;j=1,\dots ,J_M\right\}$ derived from matched scRNA-seq data from primary and metastatic sites. We organize the data such that rows represent genes, whereas cells constitute the columns. Any row or column in each count matrix that contains only all-zero values will be excluded from further analysis.

Model’s outputs

After the model training, we can obtain results at two levels. At the cell level, we identified: (i) occult metastatic cells in metastatic sites, (ii) metastatic precursor cells, which are defined as the cells that are predicted in the same cluster as the occult metastatic cell we identified in (i). At the gene level, we identified biomarkers of metastatic precursor cells for future metastasis prediction.

Heterogeneous bipartite graph construction

To formulate the cross-modal relationships between primary and metastatic sites, the matched scRNA-seq data is represented by one heterogeneous bipartite graph, comprising nodes of cells and genes, which capture these interactions and dependencies into a unified framework. In our case, we integrate matrices ${\mathrm{X}}^{\mathrm{P}}$ and $X^P$ into a combined matrix $X=\left[X^P,X^M\right]$ by constructing a gene-cell heterogeneous bipartite graph $\mathrm{G}$, consisting of two node types and one edge type to ensure each type of element (nodes and edges) maintains a unique distribution and furnishes a natural representation framework. We define the heterogeneous graph as $\mathrm{G}=\left(\mathrm{V},\mathrm{S},\mathrm{E},\mathrm{F}\right)$ with node set $\mathrm{V}={\mathrm{V}}^{\mathrm{C}}\cup {\mathrm{V}}^{\mathrm{G}}$, where ${\mathrm{V}}^{\mathrm{C}}=\left\{v_j^C\mid j=\mathrm{1,\; 2},\dots J_P,J_P+1,\dots ,J_P+J_M\right\}$ denotes all cells, ${\mathrm{V}}^{\mathrm{G}}=\left\{v_i^G\mid i=\mathrm{1,\; 2},\dots ,I\right\}$ denotes all genes. The sign set is a binary vector $\mathrm{S}=\left\{s_j\mid s_j\in {\left\{0,1\right\}}^J\right\}$. $s_j=$0 means the cell $s_j$ is in primary sites, and $s_j=$1 means the cell is in metastatic sites. The edge set $\mathrm{E}$ is constituted as $E=\left\{\left(v_i^G,v_j^C\right)\mid i=\mathrm{1,\; 2},\dots ,I,j=\mathrm{1,\; 2},\dots ,J_P+J_M\right\}$, with edge weight $w$ defined as follows. To eliminate information redundancy between node initial embeddings and the edge weights, we utilize unweighted edges when constructing the heterogeneous bipartite graph. For ${\mathrm{X}}_{i,j}>0,w\left(v_i^G,v_j^C\right)=1,$ otherwise, $w\left(v_i^G,v_j^C\right)=0$. Lastly, we establish the initial feature vectors $\mathrm{F}$ for nodes in $G$ as follows:

$$F_j^C={\mathbf{X}}_{\cdot ,j},j=1,\dots ,J_P+J_M$$ 1

$$F_i^G={\left({\mathbf{X}}_{i,\cdot }\right)}^T,i=1,\dots ,I$$ 2

where $X_{i,\cdot }$ and $X_{\cdot ,j}$ represent the $i_{th}$ row vector and the $j_{th}$ column vector of $X_{\cdot ,j}$, respectively.

EmitGCL loss design

We used the same sub-sampling strategy of heterogeneous bipartite graph and embedding update as that in MarsGT for rare cell identification. In our model training, we devise a multi-task loss function, which consists of four critical components. Loss component (1) is designed to obtain high-quality node embeddings. To reduce the false positive in the occult metastatic cells we identified, we design loss component (2) to involve prior biological information, the metastasis-related pathways, to penalize the rare cell we identified as occult metastatic cells but with low metastasis-related active score. Then, to improve the sensitivity of the occult metastatic cell identification, we design the loss component (3), a contrastive learning loss to enhance the difference between the primary site and metastatic sites, which helps us to capture slightly changes and the metastatic trail. The loss component (4) is the key to keeping the balance of the rare and major cell group signal. The formulation of the multi-task loss function is defined as follows:

$$\mathrm{Loss}={\overbrace{{\mathrm{KL}}_{\mathrm{loss}}}}^{\left(1\right)}+{\overbrace{{\mathrm{UCell}}_{\mathrm{loss}}}}^{\left(2\right)}+{\overbrace{{\mathrm{Const}}_{\mathrm{loss}}}}^{\left(3\right)}+{\overbrace{{\mathrm{Reg}}_{\mathrm{loss}}}}^{\left(4\right)}$$ 3

Where ${\mathrm{KL}}_{\mathrm{loss}}=\mathrm{KL}\left(H^l\left[V^G\right]*{H^l\left[V^C\right]}^T,\mathrm{X}\right)$. $H^l\left[V^G\right]$ and $H^l\left[V^C\right]$ means the gene and cell embedding of the $l^{\mathrm{th}}$ layer respectively. We consider occult metastatic cells as rare cell groups in the metastatic sites. Due to this property, false positives may occur. To reduce false positives, we incorporate prior knowledge of metastatic-related pathways. We then calculate the activation score of these pathways in the rare cell groups identified in the metastatic sites. The ${UCell}_{\mathrm{loss}}$ is defined as:

$${\mathrm{UCell}}_{\mathrm{loss}}=-{\sum }_{C_{\mathrm{o}}}\frac{1}{N\left(p\right)}{\sum }_p\left({\mathrm{Pathwa}{y}_{\mathrm{score}}}_p\right)$$ 4

Where $C_{\mathrm{o}}$ is a cell set which is identified as the candidate occult metastatic cells. $N\left(C_{\mathrm{o}}\right)$ is the cell number of $C_{\mathrm{o}}$, $p$ is the metastasis-related pathway we used as the prior biological information to reduce false positive of prediction. $N\left(p\right)$ is the pathway number we used. The ${{\mathrm{Pathway}}_{\mathrm{score}}}_p$ is the pathway $p$ active score, which is defined as:

$${\mathrm{Pathwa}{y}_{\mathrm{score}}}_p=\frac{1}{N\left(C_{\mathrm{o}}\right)}{\sum }_{i\in G_p,j\in C_{\mathrm{o}}}{\mathbf{X}}_{i,j}^M$$ 5

Where $G_p$ is a gene set in pathway $p$.

To depict the metastatic trail from the primary site to metastatic sites and enhance the model’s sensitivity, we designed the contrastive loss ${\mathrm{Const}}_{\mathrm{loss}}$ with three distance comparisons:

$${\mathrm{Const}}_{\mathrm{loss}}={\sum }_{\mathrm{metastatic}\mathrm{cell}\mathrm{groups}}{\mathrm{Const}}_{e{m}_{\mathrm{om}}}+{\mathrm{Const}}_{e{m}_{\mathrm{pp}}}+{\mathrm{Const}}_{p{p}_{\mathrm{op}}}$$ 6

First, consider that the distance between cells in the same occult metastatic cells should be shorter than the distance between occult metastatic cells and other cells in the metastatic sites. We set the distance between cells in the same occult metastatic cells as the positive sample pairs and the distance between occult metastatic cells and other cells in the metastatic sites as the negative sample pairs. This involves the contrastive loss term ${\mathrm{Const}}_{{\mathrm{om}}_{\mathrm{oc}}}$ which is defined as follows:

$${\mathrm{Const}}_{o{m}_{\mathrm{oc}}}=-{\sum }_{\forall {\left({\mathbf{v}}_i^C,{\mathbf{v}}_j^C\right)}_{i\ne j}\in \mathrm{om}}\log \left(\frac{e^{\frac{\mathrm{sim}\left({\mathbf{z}}_i,{\mathbf{z}}_j\right)}{\tau }}}{e^{\frac{\mathrm{sim}\left({\mathrm{z}}_i,{\mathbf{z}}_j\right)}{\tau }}+{\sum }_{{\mathbf{v}}_k^C\in \mathrm{oc}}{e}^{\frac{\mathrm{sim}\left({\mathrm{z}}_i,{\mathbf{z}}_k\right),}{\tau }}}\right)$$ 7

Where $\mathrm{om}$ is the identified occult metastatic cell set, $\mathrm{oc}$ is set of other cells in metastatic sites. ${\mathbf{z}}_i$, ${\mathbf{z}}_j$, ${\mathbf{z}}_k$ represents the embedding of cell ${\mathbf{v}}_i^C$, ${\mathbf{v}}_j^C,{\mathbf{v}}_k^C$, τ is a scaling factor used to control the concentration level of the distribution. Lower values make the model focus more on the most similar pairs, while higher values spread the focus over a larger set of pairs. $\log \left(\cdot \right)$ is the nature logarithm function, which is applied to the probability to ensure numerical stability and to bring the loss value to a suitable range. $e^{\cdot }$ means the exponential function, which is applied to the similar scores divided by the temperature to convert them into a probability-like value. $\mathrm{sim}\left({\mathbf{z}}_i,{\mathbf{z}}_j\right)$ means the similarity between cell ${\mathbf{v}}_i^C$ and cell ${\mathbf{v}}_j^C$ which defined as:

$$\mathrm{sim}\left({\mathbf{z}}_i,{\mathbf{z}}_j\right)=\frac{{\mathbf{z}}_i.{\mathbf{z}}_j}{\mid \mid {\mathbf{z}}_i\mid \mid \mid \mid {\mathbf{z}}_j\mid \mid }$$ 8

Where ⋅ denotes the dot product and $\mid \mid \cdot \mid \mid$ denotes the $L_2$ norm.

To enhance the difference between the metastatic precursor cells and the occult metastatic cells, we designed the contrastive loss ${\mathrm{Const}}_{o{m}_{\mathrm{pp}}}$ as follows:

$${\mathrm{Const}}_{{\mathrm{om}}_{\mathrm{pp}}}=-{\sum }_{\forall {\left({\mathbf{v}}_i^C,{\mathbf{v}}_j^C\right)}_{i\ne j}\in \mathrm{om}}\log \left(\frac{e^{\frac{\mathrm{sim}\left({\mathbf{z}}_i,{\mathbf{z}}_j\right)}{\tau }}}{e^{\frac{\mathrm{sim}\left({\mathrm{z}}_i,{\mathbf{z}}_j\right)}{\tau }}+{\sum }_{{\mathbf{v}}_k^C\in \mathrm{op}}{e}^{\frac{\mathrm{sim}\left({\mathrm{z}}_i,{\mathbf{z}}_k\right),}{\tau }}}\right)$$ 9

Where $om$ is the predicted occult metastatic cell set, $pp$ is the precursor cell set.

Furthermore, consider the difference between the precursor cells and other cells in primary site, we design the third contrastive loss:

$${\mathrm{Const}}_{{\mathrm{pp}}_{\mathrm{op}}}=-{\sum }_{\forall {\left({\mathbf{v}}_i^C,{\mathbf{v}}_j^C\right)}_{i\ne j}\in \mathrm{pp}}\log \left(\frac{e^{\frac{\mathrm{sim}\left({\mathbf{z}}_i,{\mathbf{z}}_j\right)}{\tau }}}{e^{\frac{\mathrm{sim}\left({\mathrm{z}}_i,{\mathbf{z}}_j\right)}{\tau }}+{\sum }_{{\mathbf{v}}_k^C\in \mathrm{op}}{e}^{\frac{\mathrm{sim}\left({\mathrm{z}}_i,{\mathbf{z}}_k\right),}{\tau }}}\right)$$ 10

Where $pp$ is the predicted occult metastatic cell set, $op$ is set of other cells in primary site.

Combining the three contrastive learning losses allows us to more sensitively identify occult metastatic cells and precursor cells, thereby depicting the tumor metastasis process.

Finally, to avoid losing all major cell information in metastatic sites, we designed a regularized term ${\mathrm{Reg}}_{\mathrm{loss}}$ as follows:

$${\mathrm{Reg}}_{\mathrm{loss}}=\mathrm{Smoothin}{g}_{\mathrm{cross}\left(P,\mathrm{L},\mathrm{\epsilon }\right)}$$ 11

where $L=\left\{l_j\mid l_j\in \left\{1,\dots ,\mathrm{T}\right\},\mathrm{j}=1,\dots ,J_P+J_M\right\}$ is the cell cluster results by Louvain with scRNA-seq from both primary and metastatic sites, $\mathbf{P}$ is the predicted cell cluster results of the model. $T$ is the number of cell clusters by Louvain, $\epsilon$ is a smoothing factor.

$${\mathrm{Smoothing}}_{\mathrm{cross}\left(P,\mathrm{L},\mathrm{\epsilon }\right)}=-{\sum }_{j=1}^N{\sum }_{t=1}^T{y}_{jt}\log p_{jt}$$ 12

$$y_{jt}=\left\{\begin{array}{c}\hfill 1-\mathrm{\epsilon }+\frac{\epsilon }{T}\mathrm{if}\left(l_j=\mathrm{t}\right)\hfill \\ \hfill \frac{\epsilon }{T}\mathrm{if}\left(l_j\ne \mathrm{t}\right)\hfill \end{array}\right)$$ 13

where $p_{jt}$ represents the predicted probability of the given cell $v_j^C$ belonging to the class $t$.

EmitGCL identified occult metastatic and precursor cells

After the training finished, we obtained the cell probability matrix $P_{J\times F}$. The row represents a cell, the column represents predicted cell groups. The value in $P_{J\times F}$ means the probability of $v_j^C$ belongs to cell group $f$. The cells will be defined as occult metastatic cells if (i) in the metastatic site, (ii) the cell number in the candidate cell group is less than 3%, (iii) the cell group with high tumor marker expression value, and (iv) the cell number of the metastatic site in the candidate cell groups is lower than that in the primary site.

The cells will be defined as precursor cells if the cells are in the same cell group with identified occult metastatic cells.

EmitGCL identified biomarkers

After the training, we obtained all cell groups of the primary and metastatic sites. For a target cell ${\mathbf{v}}_k^C$, we can obtain an attention matrix $\mathbf{A}=\left\{{\mathbf{A}}_{i,k,h}\mid i=1,2,\dots ,I,h=1,2,\dots ,H\right\}$, where the rows represent genes, and the columns represent head. The elements in the attention matrix indicate the contribution of each gene to the target cell in the cell clustering results. Considering that both negative and positive attention can contribute to distinguishing cell clusters, we incorporate the attention values across all heads in the target cell using an absolute value operation. If a gene has a high attention value across all heads, it will be identified as a gene signature of the target cell. For a targeted cell group $T$ we identified, the contribution vector $C_T$ with $I$ length represents the contribution of all genes to the targeted cell group $T$ which is defined as:

$$C_{:,T}=\frac{1}{\mid T\mid }{\sum }_{{\mathbf{v}}_k^C\in T}{\max }_{h\in \left\{1,\dots ,H\right\}}\left(\mid {\mathbf{A}}_{:,k,h}\mid \right)$$ 14

Then, we get a contribution matrix $C$ of all genes to all cluster groups. The row of the matrix represents the gene, the column represents the cell group we identified in our model.

Baseline tools parameter set

To evaluate the performance of EmitGCL relative to other tools for the FPR and sensitivity in occult metastatic cell identification, we conducted a comparative analysis between EmitGCL and other established methods.

1. MarsGT²⁵ (Python package, v 0.2.1, https://github.com/OSU-BMBL/marsgt): our in-house tool for rare cell identification, which demonstrated the best performance overall. Both the dimensionality reduction function “NodeDimensionReduction” and the prediction function “MarsGT_pred” were executed using default parameters.

2. CellSIUS²⁷ (R package, v 1.0.0, https://github.com/Novartis/CellSIUS): cells were filtered based on the total number of detected genes, total UMI counts, and the percentage of total UMI counts attributed to mitochondrial genes. Genes had to be present with at least 3 UMIs in at least one cell. After this initial QC, the remaining outlier cells were identified and removed using the “plotPCA” function from the scatter package (with detect_outliers set to TRUE). Data were normalized using the scran package, including a first clustering step as implemented in the “quickCluster” function.

3. GiniClust¹⁷ (Python package, v 3.0, https://github.com/rdong08/GiniClust3): the “neighbors” parameter of the function “clusterGini” was set to 10 (the recommended value is from 5 to 15), while other parameters were kept at their default values.

4. Seurat²⁶ (R package, v 4.3.0, https://satijalab.org/seurat): the “FindClusters” function was used for clustering with default parameters. Other functions, such as “NormalizeData,” “FindVariableFeatures,” “RunPCA,” and “RunUMAP,” were also executed using default settings.

Rareness

Defined as the proportion of occult metastatic cells relative to the total number of cells in metastatic sites.

Gene activity score

We used the AddModuleScore_Ucell function from the UCell package⁴⁷ (R package, v 2.3.1) to calculate gene activity scores for different reference gene sets. This function applies the Wilcoxon rank-sum test (U statistic) to assess the activity of gene sets by comparing the expression levels of each gene within the set to all other genes. The output is a relative activity score computed for each cell.

EMT activity score

To calculate the EMT activity score, we retrieved the HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION gene set from the MSigDB database using the msigdbr package (R package, v 7.5.1). A high EMT score indicates a mesenchymal-like phenotype, which is often associated with cancer metastasis. In contrast, a low EMT score corresponds to an epithelial-like phenotype, typically related to cell adhesion and a reduced likelihood of migration and invasion.

CNV score

CNV scores were computed using the “tl.infercnv” function of the infercnvpy package (Python package,v 0.4.2, https://github.com/icbi-lab/infercnvpy). The core idea is to compare single-cell RNA sequencing data against reference normal cell populations to identify gene expression changes indicative of CNV. This method helps in detecting large-scale chromosomal alterations by assessing expression deviations.

Tumor stemness

To calculate the tumor stemness score, we collected literature on stemness marker genes for various types of cancer and calculated the gene activity score to represent tumor stemness^{28,34,48–51} (Supplementary Data 20). Cells with higher scores are more likely to be occult metastatic cells.

Metastatic enrichment score

To calculate the metastatic enrichment score, we select the metastasis-related pathway in Gene Ontology (GO) terms and calculated the gene activity score to represent tumor stemness. Cells with higher scores are more likely to be metastatic cells.

Overview score

In each dataset used for computational evaluation, we rank five tools across EMT, CNV and tumor stemness indexes. A higher rank corresponds to a higher score. The highest possible score in a single dataset is 15, while the lowest is 3.

True positive rate (TPR)

$TPR$ refers to the proportion of actual positive cases (patients with metastasis) that are correctly identified by the model. It is defined as follows:

$$\mathrm{TPR}=\frac{\mathrm{TP}}{\mathrm{TP}+\mathrm{FN}}$$ 15

where:

• $TP$ (true positive) is the number of patients who have metastasis and are correctly classified as having metastasis.

• $\mathrm{FN}$ (false negative) is the number of patients who have metastasis but are incorrectly classified as not having metastasis.

False positive rate (FPR)

$FPR$ refers to the proportion of actual negative cases (patients without metastasis) that are incorrectly classified as positive by the model. It is defined as follows:

$$\mathrm{FPR}=\frac{\mathrm{FP}}{\mathrm{FP}+\mathrm{TN}}$$ 16

where:

• $FP$ (false positive) is the number of patients who do not have metastasis but are incorrectly classified as having metastasis.

• $\mathrm{TN}$ (true negative) is the number of patients who do not have metastasis and are correctly classified as not having metastasis.

Trajectory analysis

We used the “orderCells” function from the Monocle2⁵² (R package, v2.26.0) to infer cell developmental trajectories and rank the cells accordingly.

Simulated TF knockout

We utilized the CellOracle⁴⁴ (Python package, v0.12.0) to perform simulated transcription factor TF knockout analyses. First, we constructed gene regulatory networks using the “celloracle.data.load_human_promoter_base_GRN” function, which provides pre-built promoter base GRNs for human cells. Next, we applied the “celloracle.simulate_shift” function to simulate the knockout of specific TFs. This function recalculates the predicted gene expression profiles by removing the regulatory influence of the selected TFs, allowing us to assess the potential impact of the TF knockout on downstream gene expression and cellular behavior.

Statistics and reproducibility

Statistics: all statistical analyses were performed in R version 4.2.1 (ggpubr 0.6.0, survival 3.8.3, and survminer 0.5.0) and Python version 3.8.0 (scipy 1.9.1). All boxplots display the median as the central line, the interquartile range (IQR; 25th–75th percentile) as the box and outliers (points beyond 1.5 times the IQR) as dots outside the whiskers. Both line plots with error bands and bar plots with error bars display mean values, with the error bands and error bars representing the standard deviation above and below the means.

Experimental: the transwell migration assay was performed with five to six biological replicates per group (control, n = 5; each knockout group, n = 6). All experiments yielded consistent results, demonstrating reproducibility. No statistical method was used to predetermine sample size for the in vitro experiments. A power calculation for the in vivo study, based on the in vitro migration results, indicated that four to six mice per group would detect a statistically meaningful difference, and six mice per group were used. No data were excluded from the analyses. The experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment.

Model: to evaluate the stability and reproducibility of EmitGCL, we randomly selected a dataset from each cancer type and then repeated the algorithm 20 times using the default parameters. Finally, we calculated the ARI between the different runs for each selected dataset within each cancer type to assess the consistency of the clustering results. To evaluate robustness to random initialization, we tested five datasets (one randomly selected per cancer type), each run with 20 different random seeds. The seed used in the main text served as the reference. For each alternative seed, we computed the ARI and normalized mutual information (NMI) relative to reference clustering.

Clinical data integration

To integrate the data, we first collected the intersection of genes from five batches of uncorrected expression data and then filtered out genes were not present in the intersection across all datasets. We then integrated batch information from the different datasets into a single AnnData object. The following clinical data were added to the combined dataset: batch information (batch: Chin, Caldas, Desmedt, Miller, and TCGA), time information (time), and label information (label: high metastatic potential (time ≤ 5), low metastatic potential (time > 5)).

Clinical differential gene screening

Differentially expressed genes (DEGs) were identified within each batch using the Wilcoxon rank-sum test based on the label. For each batch, genes with p < 0.05 and logFC > 0 were selected as DEGs. The significant genes were then divided according to their batch, and an UpSet plot was generated to identify common DEGs across multiple batches. Common genes included GATA3 (shared by Caldas and Chin) and AZGP1 (shared by Caldas and Desmedt). Finally, GATA3 and AZGP1 were selected as the DEGs for benchmarking.

Bulk RNA-seq batch effect correction

Batch effects in the integrated data were corrected using the ComBat⁵³ method, resulting in corrected expression data for further analysis. We set parameters as follows:

• i)key = 'batch': ['Chin','Miller','Desmedt','TCGA','Caldas'], capturing study/platform-level effects.
• ii)group (covariates): not included to avoid regressing out biologically relevant differences related to outcome/subtype and to retain signals useful for prediction.

Model feature selection

The expression data of the corrected DEGs, GATA3 and AZGP1, were extracted for model feature validation. Models were established using these feature data to analyze the relationship between genes in the training set. The model’s accuracy and predictive performance were evaluated across different gene combinations using random seeds for training. To visualize the results, we plotted the mean and maximum accuracy of each gene combination and created Kaplan–Meier survival curves to analyze the survival status of patients with early or late metastasis based on gene combinations. Finally, HSP90AA1 and HSP90AB1 were identified as the final model features.

Classification model performance evaluation

The expression data of the corrected DEGs, GATA3 and AZGP1, were used to evaluate the classification model’s performance. We focused on samples with late metastasis (time > 5) for GATA3 and AZGP1 expression, using various random seeds to assess classification accuracy. The scatter plot of GATA3 and AZGP1 gene expression was visualized, and the corresponding classification boundaries were drawn. The classification performance was illustrated based on the identified features for samples in early and late metastasis stages.

Clustering model performance evaluation

For clustering model performance evaluation, we extracted the expression data of the corrected DEGs, GATA3 and AZGP1. K-means clustering was employed to group the data into two clusters. The clustering accuracy was calculated, and the results were visualized. The clustering results were displayed, and the distribution of clusters based on gene expression was analyzed.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Supplementary information

Supplementary Information ^{(1.6MB, pdf)}

41467_2026_76277_MOESM2_ESM.pdf ^{(190.8KB, pdf)}

Description of Additional Supplementary Files

Supplementary Data 1–20 ^{(547.4KB, xlsx)}

Reporting Summary ^{(1.9MB, pdf)}

Transparent Peer Review File ^{(2.6MB, pdf)}

Source data

Source Data ^{(887.8KB, xlsx)}

Author contributions

Q.M. conceived the basic idea. R.L.C. and A.J.S. performed in vitro and in vivo experimental validation. X.W. designed the algorithm, conducted the case study, and wrote the manuscript. M.D. carried out benchmark experiments and wrapped the code. P.S. performed survival analysis. J.L., J.J., Y.X., and G.X. designed and conducted the cell line study. J.K. provided biological insights for this study. P.S., K.H., Z.L., and D.P.C. provided clinical rationale for this study. D.X. provided deep learning knowledge. H.C., Y.S., and W.W. contributed to methodological guidance and discussion interpretation. G.W., C.Z., and S.C. contributed expert insights during manuscript preparation and provided critical feedback. L.L. and Y.X. contributed to the revision. W.C. supported spatial-related analysis. J.Z. contributed to statistical rigor and reproducible validation.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by research grants R01GM152585 (Q.M.), P01CA278732 (Z.L. and Q.M.), and R01CA285967 (R.L.C.) from the National Institutes of Health. This work was supported by the Pelotonia Institute of Immuno-Oncology (PIIO).

Data availability

Previously published scRNA-seq datasets were used for benchmarking and case studies in this study. Breast cancer scRNA-seq datasets were obtained from GEO under accession numbers GSE167036 and GSE180286. Lung cancer datasets were obtained from GEO under accession number GSE198099. Head and neck cancer datasets were obtained from GEO under accession number GSE188737. Pancreatic cancer datasets were obtained from GEO under accession number GSE197177. Papillary thyroid cancer datasets were obtained from GEO under accession number GSE241184. Nasopharyngeal carcinoma scRNA-seq datasets were obtained from the GSA for Human database under accession number HRA000036. Additional previously published RNA-seq datasets from UCSC were used for breast cancer studies, including datasets from Chin (2006), Miller (2005), Desmedt (2007), Wang (2005), and Caldas (2007) [https://xenabrowser.net/datapages/]. All datasets were reprocessed and analyzed as described in the Methods section. Details of data information can be found in Supplementary Data 1. Source data are provided with this paper.

Code availability

The EmitGCL source code used in this study is publicly available (source-available) on GitHub (https://github.com/mtduan/emitgcl) and archived on Zenodo⁵⁴ under the PolyForm Noncommercial License 1.0.0, which permits non-commercial use while the related patent application (PCT Publication No. WO2026112293; applicant: Ohio State Innovation Foundation) is pending.

Competing interests

Q.M. and X.W. are inventors on a patent application related to the methods described in this paper (PCT International Application No. PCT/US2025/056344, International Publication No. WO2026112293, filed 20 November 2025 by Ohio State Innovation Foundation; title: Systems and Methods for Metastatic Cell Detection and Risk Prediction). The remaining authors declare no competing interests.

Supplementary information

The online version contains Supplementary material available at https://doi.org/10.1038/s41467-026-76277-x.