Figure 3.

Spatial location inference across platforms

(A) Evaluation of spatial reconstruction for the mouse gastrulation cells using array-based ST reference. The training dataset consistently utilized the Stereo-seq E2S1 sample. The benchmarking was against three independent single-cell-resolution ST datasets derived from E8.5 mouse embryos.

(B) Evaluation of cell neighbor identification. We evaluated the efficacy of different methods by quantifying the average hit number within varying k nearest neighbors across the three distinct mouse test datasets; the mean average hit numbers of three testing sets are presented.

(C) Evaluation of local neighborhood cell-type heterogeneity reconstruction. We computed the average JSD for each cell type across the three mouse test datasets and showed the mean values of the three testing sets.

(D) Evaluation of spatial reconstruction for 10X Visium human breast cells. The training utilized the normal sample S06, consisting of 3,037 spots, while the testing involved sample S08, comprising 2,801 spots.

(E) Assessment of cell neighbor identification. We quantified performance by analyzing the average hit number across varying nearest k neighbors of the test human breast cells.

(F) Evaluation of neighborhood cell-type reconstruction of human breast cells using JSD.