Table 3.
The characteristics of reviewed papers entered in the study based on inclusion criteria.
| # | Author | Journal/ Conference name | Research aim | Cancer Type | Kind of care (Prediction, Screening, Diagnosis, Treatment) | AI method | Categories of AI approaches | Input | Output | Software | Source of Data set | Sample Size # |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Le and Pham [43], Vietnam | Journal of Molecular Biology | The main aim of this work was developing a novel network-based method, named GloNetDRP, to overcome response limitation | More than one type of cancer | Predict drug response | GloNetDRP | Linear model | Genomic | Cell line similarity and drug similarity for predict response of to drug | Not mentioned | CCLE + GDSC | Not mentioned |
|
| ||||||||||||
| 2 | Gatta et al. [44], Italy | Artificial Intelligence in Medicine | The objective of this paper was introducing a medical agent-based decision support system capable of handling the whole radiomics process and return a prediction about the clinical outcome of a proposed treatment | Rectal cancer | Prediction of cancer from samples | LR+ FE | Rule-based system | Radiomics | Predicting the outcome of a previously unseen clinical case | RadAgent software | To provide the appropriate input for the proposed approach, MR scans have been processed using the moddicom R library | Data set 1: 173 patients Data set 2: 25 patients |
|
| ||||||||||||
| 3 | Shimizu and Nakayama [45], Japan | EBioMedicine | The aim of this work was developing a new prognostic score based on intelligence models (random forest and neural network) that is enforceable to a wide range of patients with breast cancer | Breast cancer | Prognosis prediction and scoring stage | RF+ NN | Linear and nonlinear models | Genomic | Overall Survival (OS) or clinical stage and disease-free survival (DFS) | Not mentioned | TCGA + molecular taxonomy of Breast Cancer International Consortium | Patients: 11,893 |
|
| ||||||||||||
| 4 | Sun et al. [25] China | EBioMedicine | The main aim was constructing a predictive model for predicting the response to neoadjuvant chemotherapy (NACT) by radiomic analysis | Cervical cancer | Predict drug response | SVM | Linear model | Radiomics | “responders” (chemosensitive); “nonresponders” (chemoinsensitive) | MATLAB | No public data: Nanfang hospital | Total 275 patients (183 patients training sets; 92 patients testing sets) |
| 5 | Overby et al. [46], USA | The Journal of the American Medical Informatics Association | This study was presenting a knowledge-based approach to derive “phenotype score” based on Pharmacogenomics knowledge. This model has prediction power for metabolic process in drug treatments (to breast cancer patients taking tamoxifen | Breast cancer | Predict drug response | PEMRIC | Rule-based system | Drug response-Functional data | Endoxifen/ NDM plasma levels | Not mentioned | Clinical data source; Pharmacogenomics Knowledge Base and evidence-Base (PharmGKB, SuperCYP) | Patients: 30 |
|
| ||||||||||||
| 6 | Urbanowicz et al. [47], USA | The Journal of the American Medical Informatics Association | The aim of this work was detecting complex patterns of association between genetic on environmental risk factors | Bladder cancer | Diagnosis | AF-UCS | Non-linear model | Genomic | Tumor stage and grade, age at diagnosis (years), survival time in years | Graph visualization software | Source of data set was provided by Andrews et al. work. | Case group: 355 control group: 559 |
|
| ||||||||||||
| 7 | Ding MQ et al. [48], USA | Molecular Cancer Research | The aim of this study was to identify drugs that will be effective treat neoplasms | More than one type of cancer | Predict drug response | SVM | Linear model | Drug response-Functional data | The name of targeted drugs and percentage of cell line responsive | Not mentioned | GDSC + CCLE | The 624 cell lines were randomly split into training and testing data sets of 520 and 104 samples, |
|
| ||||||||||||
| 8 | Diggans et al. [49], USA | Pacific Symposium on Biocomputing | In this article, the main aim was developing a machine learning approach for identifying the BRAF V600E mutations using MRNA expressions in thyroid fine needle aspirate biopsies (FNABs) | Thyroid neoplasm | Diagnosis | SVM | Linear model | Genomic | Malignant nodules and benign nodules | Not mentioned | Not mentioned | Patients: 716 |
| 9 | Gligorijevic et al. [50], UK | Pacific Symposium on Biocomputing | In this work, the main aim was introducing a versatile data fusion framework that was based on graph-regularized nonnegative matrix trifactorization, a machine learning technique for co-clustering heterogeneous datasets | More than one type of cancer | Predict drug response | NMTF | Linear model | Genomic | Patient subgroups (stratification) with prognostic survival outcome, predicting novel driver genes and repurposing drugs (predicting new candidate drugs) | Not mentioned | TCGA4 with molecular networks (MNs) from BioGRID17 and KEGG18, drug-target interaction (DTI) and DrugBank | Patients: 353 |
|
| ||||||||||||
| 10 | Acar et al. [51], Turkey | The British Journal of Radiology | In this study, the team used different machine learning methods for distinguish the lesions images | Prostate cancer | Treatment | DT+ SVM+ KNN+EC | Linear and non-linear models | Radiomics | Metastasis/Completely responded lesions | LifeX software-MATLAB software | Medical records | Patients: 75 |
|
| ||||||||||||
| 11 | Alaa AM et al. [52], USA | IEEE Transactions on multimedia | The objective of this work was presenting a CDSS for stratifying cluster of patients | Breast cancer | Screening | ConfidentCare | Rule-based system | Radiomics | Recommend a regular (1 year) follow-up, recommend a diagnostic test (biopsy) | Not mentioned | EHR data in the United States | Patients: 25,594 |
|
| ||||||||||||
| 12 | Boucheham and Batouche [53], Algeria | Science and Information Conference | The main aim was proposing a novel algorithm for biomarker discovery in cancer diagnosing | Colon, leukemia and, ovarian | Diagnosis | MEFS | Linear model | Genomic | The best selected features | MATLAB | Kent Ridge biomedical data repository | Patients with colon cancers: 62 leukemia: 72 ovarian cancers: 253 |
|
| ||||||||||||
| 13 | Breitenstein et al. [26], USA | Clinical and Translational Science | The main aim was proposing a rule-based algorithm which could create robust precision medicine phenotypes in breast cancer patients from HER perspectives | Breast cancer | Diagnosis + treatment | NLP | NLP | Genomic | Receptor status phenotypes: overexpression in BC patients | Not mentioned | Cancer registry data + EHR data. | Patients: 13,162 |
| 14 | Nikolova et al. [54], USA | Bioinformatics | The aim was proposing a novel, biologically motivated, Bayesian multitask approach, which explicitly models gene-centric dependencies across multiple and distinct genomic platforms for the identification of drug response biomarkers | More than one type of cancer | Predict drug response | GBGFA | Bayesian model | Drug response-functional data | Drug recommend based on cell lines patients | Not mentioned | CCLE + CTRP + TCGA PAAD+ LUAD cohorts | Biosamples: 267 cell lines + 409 cell lines patients: 132+ 165 |
|
| ||||||||||||
| 15 | He et al. [55], Switzerland | Bioinformatics | The aim of this work was proposing a ML approach which named Kernelized Rank Learning. This method ranks drugs based on patient's molecular profile | Breast cancer | Treatment | KRL | Linear and nonlinear models | Drug response-functional data | Drug recommend based on cell lines patients | Python + MATLAB | GDSC + TCGA | Not mentioned |
|
| ||||||||||||
| 16 | Mobadersany et al. [56] USA | Proceedings of the National Academy of Sciences of the United States of America | The aim of this work was developing a computational method based on deep learning for predicting the outcome of patients with brain tumor | Brain tumor | Predicting the clinical outcome and survival prediction | CNN | Deep learning model | Radiomics | Grading diffuse gliomas and suggest relevance for patterns with prognostic significance | TensorFlow | TCGA Lower-Grade Glioma (LGG) and Glioblastoma (GBM) projects. | Patients: 769 |
|
| ||||||||||||
| 17 | Fathiamini et al. [57], USA | Journal of American Medical Informatics Association | The objective of this work was creating an automated system to identify drugs for cancer related genes in relevant literature | More than one type of cancer | Predict drug response | NLP | NLP | Drug response-Functional data | Detect relation between gene-drugs for treating cancers in clinical trials. | Not mentioned | SemMedDB: SemRep_UTH to process MEDLINE and ClinicalTrials.gov | 183 260 trials (entire set) + 23 537 576 PubMed abstracts |
| 18 | Itahashi et al. [58], Japan | Frontiers in Medicine | The basic objective of this work was to assess the validity and utility of WfG for analyzing clinical genome sequencing results by comparisons with results obtained by an expert panel composed of multidisciplinary specialists at NCCH | More than one type of cancer | Diagnosis + treatment | WfG | Rule-based system | Genomic | Actionable or alterations with therapies | IBM Watson for Genomics | Hospital (TOP-GEAR PROJECT) | Patients: 198 |
|
| ||||||||||||
| 19 | Chen et al. [27] China | Frontiers in Medicine | The main objective was exploring a radiomic model for preoperative prediction of ETE in patients with PTC | Papillary thyroid carcinoma | Preoperative prediction | LR+ RF+ SVM | Linear and nonlinear models | Radiomics | Preoperative prediction of ETE | MATLAB | Dataset was recruited for patients in cohort study. | Patients: 624 |
|
| ||||||||||||
| 20 | Wang et al. [59], China | Frontiers in Medicine | The main objective was exploring and developing a new approach based on eight radiomic features for identifying the individuals' accurate preoperative T category for patients with advanced malignant laryngeal carcinoma | Laryngeal carcinoma | Preoperative prediction | SVM | Linear model | Radiomics | Preoperative T category (T3 vs. T4) for patients with advanced laryngeal cancer before surgery | Pyradiomics, Python + R software | Medical records | Patients: 211 |
|
| ||||||||||||
| 21 | Graim et al. [60], USA | Pacific Symposium on Biocomputing | The main aim of this work was proposing a multiple-view learning predictive framework for identifying the cancer drug sensitivity | More than one type of cancer | Predict drug response | MVL | Rule-based system | Drug response-functional data | Predicting drug sensitivity in cell lines | PLATYPUS | CCLE | Biosamples:1,037 |
| 22 | Jones et al. [61], USA | BMC Medical Genomics | The main aim of this work was identifying reliable gene expression pattern for classifying tumor class using a local minimax kernel algorithm | Leukemia and prostate cancer | Prognosis prediction | KRL | Linear and nonlinear models | Genomic | Predicting the probability of malignancy with a level of confidence-diagnose |
Not mentioned | Three publicly available gene expression datasets: data was extracted from papers | Tumor samples: 365 normal samples: 265 |
|
| ||||||||||||
| 23 | Kim et al., USA | IEEE International Conference on Bioinformatics and Biomedicine (BIBM) | The main aim of this work was proposing a framework based on personalized medicine with Reverse-Phase Protein Array (RPPA) and sensitivity of drugs | Lung cancer | Predict drug response | NB | Bayesian model | Genomic | High probability of low sensitivity or low probability of low sensitivity | Weka | Not mentioned | Biosamples: 55 antibodies and 75 lung cancer cells lines, cell lines per drug is 43 |
|
| ||||||||||||
| 24 | Kureshi et al. [62], Canada | IEEE Journal of Biomedical and Health Informatics | The main objective of this work was investigating the influence of a combination of factors-clinical predicators, environmental risk factors, and EGFR mutation. These can be used to predict the tumor response to EGFR-TKI therapy for patients with advanced-stage NSCLC | Lung cancer | Treatment | SVM + DT | Linear and nonlinear models | Genomic | Responder and nonresponder group (response to EGFR-TKI therapy.) | Weka | PubMed papers | Train set: 291 patients Test set: 64 patients |
|
| ||||||||||||
| 25 | Li et al. [63], China | IEEE International Conference on Bioinformatics and Biomedicine (BIBM) | The main objective was developing a centric radiogenomics framework based on a deep learning approach for mapping the image features and characteristics and gene expression profile data | Lung cancer | Diagnosis | CNN | Deep learning model | Radiomics | Image features and patients' metagene: typical CT TR and gen information | Not mentioned | Dataset lung3 and NSCLC Radiogenomics are from cancer archive |
Patients: 300 |
| 26 | Lin et al. [64], China | European Radiology | The aim of this work was to develop a radiomics and genomic signature to predict clinical outcomes and prognosis of BLCA patients | Bladder cancer | Predicting the clinical outcome and survival prediction | LASSO | Nonlinear model | Radiomics | Prognostic indicators: High-risk or Low-risk | R software + ultrosomics software | TCGA + TCIA | Patients: 62 |
|
| ||||||||||||
| 27 | Menden et al. [65], UK | PLoS ONE | The main objective of this paper was to develop a machine learning model to predict sensitivity and drug responses based on genomic features and alterations | More than one type of cancer | Predict drug response | MLP + RF | Linear and nonlinear models | Drug response-Functional data | Cell survival and drug response | Encog + R + PaDELe | Genomics of Drug Sensitivity in cancer project | Cell line samples: 608 drugs:111 |
|
| ||||||||||||
| 28 | Moon et al. [66], USA | Artificial Intelligence in Medicine | The main objective of this work was proposing a new ensemble-based classification method than can be used to predict more effective therapies in patients for individualizing treatments | Lymphoma, lung cancer, and breast cancer | Predict drug response | CART | Linear model | Drug response-functional data | Classifying patients and drug response | R package | Public websites | Patients: 306 |
|
| ||||||||||||
| 29 | Assawamakin et al. [67], Thailand | BioMed Research International | The basic aim of this paper was developing a novel two-step machine learning framework which can present to address the prediction of phenotypic outcomes. | More than one type of cancer | Prognosis prediction | NB + HNB | Bayesian model | Genomic | Prediction of phenotypic and proteomic outcomes | Weka | KRBDSR + GEMLeR + NCICPD | Patients: 230 |
|
| ||||||||||||
| 30 | Majumder et al. [68], India | Nature Communications | The aim of this work was proposing a machine learning algorithm to predicting clinical response to anticancer drugs for engineering of personalized tumor ecosystems | Head and neck carcinoma and colorectal cancer | Predict drug response | SVM | Linear model | Drug response-Functional data | Ranking patients as CR (complete response), PR (partial response) or NR (nonresponse): These ranking had different drug regimens D1,D2, D3 or D4. | PSPEP Software | Dataset was recruited by project team. | Patients: 164 |
| 31 | Sun et al. [69], China | Cancers | The main objective of this study was proposing artificial intelligence approach that could predict assessments of the level of BAP1 expression in enucleated eyes with unveil melanoma | Uveal melanoma | Cancer classification | DenseNet-121 | Deep learning model | Genomic | Prediction of BAP1 classification: Yellow areas correspond to BAP1-classification “high” and green to “low”. | PyTorch toolkit + Python | Published papers | Patients: 47 |
|
| ||||||||||||
| 32 | Potie et al. [70], Spain | 2019 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE) | The objective of this study was to show the benefits of one of the learning paradigms of Computational Intelligence | Lung cancer | Cancer classification | Fuzzy model | Deep learning model | Genomic | Lung cancer prediction from samples taken by liquid biopsy. | KEEL + scikit-learn | GEO | Patients:779 |
|
| ||||||||||||
| 33 | Yu et al. [71], USA | Journal of Proteome Research | The objective of this work was predicting individual platinum response using robust machine learning models, and discovered proteins and biological processes associated with platinum response | Ovarian cancer | Treatment | RF+ SVM+NB | Bayesian model + Linear and nonlinear models | Genomic | Patient's response to platinum drugs | R package | TCGA + CPTAC | Patients: 130 |
|
| ||||||||||||
| 34 | Kalari et al. [72], USA | JCO Clinical Cancer Informatics | We propose a precision medicine computational framework, PANOPLY (Precision Cancer Genomic Report: Single Sample Inventory), to identify and prioritize drug targets and cancer therapy regimens | Breast cancer | Treatment | RF | Linear and nonlinear models | Drug response-Functional data | Personalized list of prioritized drugs | R package | Breast Cancer Genome Guided Therapy Study (BEAUTY) | Patients: 17 |
| 35 | Klein et al. [73], USA | BMC Bioinformatics | The aim of this study was presenting GRAPE as a novel method to identify abnormal pathways in individual samples that is robust to platform/batch effects in gene expression profiles generated by multiple platforms | Breast cancer | Diagnosis | RF + SVM | Linear and nonlinear models | Genomic | Use healthy reference samples to quantify the abnormality of individual pathological samples. | R package | TCGA | Different sample size mentioned for different pathways |
|
| ||||||||||||
| 36 | Dong et al. [74], China | BMC Cancer | The aim was the generation of genetic predictions of drug response in the preclinical setting and their incorporation into cancer clinical trial design | Not mentioned | Predict drug response | SVM | Linear model | Drug response-Functional data | Response to anticancer drugs | R package | CCLE + CGP | Not mentioned |
|
| ||||||||||||
| 37 | Kempowsky-Hamon et al. [29], France | BMC Medical Genomics | The study aimed to develop a new gene selection method based on a fuzzy logic selection and classification algorithm, and to validate the gene signatures obtained on breast cancer patient cohorts | Breast cancer | Prognosis prediction | Fuzzy model | Rule-based system | Genomic | We confirmed the use of fuzzy logic selection as a new tool to identify gene signatures with good reliability and increased classification power | R package | NKI2-Agilent + KJX64KJ125-GSE2990 + Uppsala-GSE4922 + Transbig-GSE7390 | Patients: 452 |
|
| ||||||||||||
| 38 | Yasser et al. [75], USA | BMC Medical Genomics | The aim was proposing a novel framework for multiomics data integration using multiview feature selection | Ovarian cancer | Predicting the clinical outcome and survival prediction | RF+ XGB +LR | Linear and nonlinear models | Genomic | Cancer survival prediction (short-term versus long-term survival) | Python | TCGA | Not mentioned |
| 39 | Chang et al. [76], South Korea | Scientific Reports | In this study, we have developed the Cancer Drug Response profile scan (CDRscan), a cancer genomic landscape-guided drug response prediction algorithm | More than one type of cancer | Treatment | CNN | Deep learning model | Drug response-functional data | Anticancer drug responsiveness | TensorFlow, Keras | CCLP + GDSC | Biosamples: Train set:144,953 Test set:7,641 |
|
| ||||||||||||
| 40 | Huang et al. [77], USA | Scientific Reports | The aim of this study was evaluating the performance of an approach to predict individual patient responses to drugs based on gene expression profiles of each individual's tumor | Ovarian cancer | Predict drug response | SVM | Linear model | Drug response-Functional data | Predicts individual cancer patient responses to chemotherapeutic drugs | Not mentioned | TCGA | Patients: 175 |
|
| ||||||||||||
| 41 | Shen et al. [28], China | IEEE Access | The aim was coming up with a new classification model named OFSSVM for cancer prediction using gene expression data | More than one type of cancer | Prognosis prediction | SVM | Linear model | Genomic | Multiclass cancer diagnosis | MATLAB | Prostate tumor dataset, AML/ALL dataset, GCM dataset | Not mentioned |
|
| ||||||||||||
| 42 | Ow et al. [78], Singapore | Scientific Reports | The objective of this work was investigating the predictive performance of PSVM via optimization of the prognostic variable weights | Ovarian cancer | Prognosis prediction | SVM | Linear model | Genomic | Three survival-significant risk groups (low-, intermediate- or high- risk) | Python | TCGA + GSE9899 + GSE26712 | Train: 349 patients Test: 359 patients |
| 43 | Xiao et al. [18], USA | Clinical Cancer Research | The aim was proposing a RWRF model, which updates the weight of each decision tree whenever additional patients` information is available, to account for the potential heterogeneity between training and testing data | Lung cancer | Predict drug response | RF | Linear and nonlinear models | Genomic | Predict the clinical response to gefitinib treatment | Not mentioned | Not mentioned | Test set: 59 patients genes: 1,473 |
|
| ||||||||||||
| 44 | Yu et al. [79], China | BMC Cancer | The aim of this study was providing a prediction model on the prognosis of lung adenocarcinomas based on somatic mutational features | Lung cancer | Prognosis prediction | SVM | Linear model | Genomic | Good and poor prognosis group | R package | TCGA | Patients:371 |
|
| ||||||||||||
| 45 | Dercle et al. [80], USA | European Journal of Radiology | Researchers aimed to develop a machine-learning algorithm for Quality Control of Contrast-Enhancement on CT-scan (CECTQC) | Liver cancer | Diagnosis + Treatment | RF | Linear model | Radiomics | Five contrast-enhancement phases in abdominal CT scan image | Python+ MATLAB+SPSS | Independent cohorts | Patients: 503 |
|
| ||||||||||||
| 46 | Zhang et al. [81], China | Radiotherapy and Oncology | A novel deep learning model was proposed to predict the risk for overall survival based on computed tomography images | Gastric cancer | Predict drug response | DL | Deep learning model | Radiomics | Risk prediction of overall survival | Not mentioned | Independent medical centers | Patients: 640 |
|
| ||||||||||||
| 47 | Su et al. [82], China | Methods | Researchers proposed a deep cascaded forest model, Deep-Resp-Forest, to classify the anticancer drug response as “sensitive” or “resistant” | More than one type of cancer | Predict drug response | Deep-Resp-Forest | Deep learning model | Genomic | Drug response prediction | Not mentioned | CCLE + GDSC | 33 to 275 cancer cells lines + 12 to 156 cell lines |
| 48 | Mahmood et al. [83], South Korea | Journal of Personalized Medicine | Researchers proposed an AI-based nuclear segmentation technique which is empowered by residual skip connections to address this issue | More than one type of cancer | Prognosis prediction | CNN | Deep learning model | Genomic | Determination of cell phenotype, nuclear morphometrics, cell classification | MATLAB | TCGA + TNBC | Patients: 20,000 |
|
| ||||||||||||
| 49 | Malik et al. [84], India | BMC Genomics | The researchers proposed a multiomics integrative framework that robustly quantifies survival and drug response for breast cancer patients | Breast cancer | Prognosis prediction | NN | Deep learning model | Multiomics | Survival and drug response: two survival classes – high-risk and low risk | R package | TCGA+ GDSC | Patients: 6221 |
|
| ||||||||||||
| 50 | Nascimento et al. [85], Brazil | BMC Medical Informatics and Decision Making | A decision tree modeling was proposed to improve the accuracy of the pathogenicity identification process | More than one type of cancer | Treatment | DT | Nonlinear model | Genomic | Genetic variant impact prediction | Not mentioned | ClinVar | 25.052 nonsynonymous mutations |
|
| ||||||||||||
| 51 | Choi et al. [86], South Korea | Scientific reports | Researchers developed a novel Reference Drug-based Neural Network (RefDNN) model for effective prediction of anticancer drug response and identification of biomarkers contributing to drug resistance. | More than one type of cancer | Predict drug response | RefDNN | Deep learning model | Genomic + Drug Response-Functional data | Response to anticancer drugs | R package | GDSC + CCLE | 1,065 cancer cell lines+ 983 cancer cell lines |
|
| ||||||||||||
| 52 | Koras et al. [87], USA | Scientific Reports | Researchers compare standard, data-driven feature selection approaches to feature selection driven by prior knowledge of drug targets, target pathways, and gene expression signatures | More than one type of cancer | Predict drug response | LR+ RF | Linear and nonlinear models | Genomic + Drug Response-Functional data | Response to anticancer drugs | Python | GDSC | 983 cancer cell lines |
| 53 | Zhu et al. [31], USA | Scientific Reports | Researchers investigate the power of transfer learning for three drug response prediction applications including drug repurposing, precision oncology, and new drug development | More than one type of cancer | Predict drug response | Elastic Net, RF, SVM |
Linear and nonlinear models | Genomic + Drug Response-Functional data | Response to anticancer drugs | Not mentioned | GDSC+CCLE | 1927 genes |
|
| ||||||||||||
| 54 | Luo et al. [88], China | Pharmacological Research | Researchers propose a computationally efficient and cost-effective collaborative filtering method with ensemble learning to shorten the decision-making process regarding the selection of the most suitable compounds for patients | Lung cancer | Predict drug response | ECF-S + ECF-W | Linear and nonlinear models | Drug response-Functional data | Response to anticancer drugs | Not mentioned | Local dataset | Eight NSCLC (nonsmall cell lung cancer) cell lines |
|
| ||||||||||||
| 55 | Maros et al. [30] Germany | Nature Protocols | Main aim was to perform a benchmark analysis to support the choice for optimal DNA methylation microarray data analysis through extensive comparisons of well-established ML classifiers such as RFs, ELNET, SVMs | More than one type of cancer | Treatment | RF+ ELNET + SVM | Linear and nonlinear models | Genomic | Classifying patients | R package + Python | TCGA | Patients: 2,801 |
|
| ||||||||||||
| 56 | Sharifi-Noghabi et al. [89], Canada | Bioinformatics | The main aim was predicting response to a drug given some single—or multiomics data | More than one type of cancer | Predict drug response | AITL | Deep learning model | Drug response-Functional data | Predict drug response | PyTorch | GDSC + TCGA+Patient-Derived Xenograft (PDX) Encyclopedia dataset+Patient datasets from nine clinical trial cohorts | Patient:618 targeted and chemotherapy drugs:299 |
| 57 | Bazgir et al. [90], USA | Bioinformatics | The main aim of this study was anticancer drug sensitivity prediction using deep learning models for individual cell line | Not mentioned | Predict drug response | REFINED-CNN | Deep learning model | Drug response-Functional data | Anticancer drug sensitivity prediction | PaDEL | NCI60 + NCI-ALMANAC databases | 17 cell lines |
|
| ||||||||||||
| 58 | Woo et al. [91], Canada | Bioinformatics | The main aim is identification of a drug candidate causing a desired gene expression response, without utilizing any information on its interactions with protein target(s) | More than one type of cancer | Predict drug response | MLP | Deep learning model | Genomic + Drug Response-Functional data | Direct identification of a drug candidate causing a desired gene expression response | Python +R | LINCS CMap L1000 cancer genomic dataset | The gene profiles of 978 landmark genes |
|
| ||||||||||||
| 59 | Jacobs et al. [92], USA | Cancers | The main aim of this study was identifying potential risk of local or systemic recurrence in breast cancer patients | Breast cancer | Prognosis prediction | SVM | Linear and nonlinear models | Radiomics | Classification of patients into different risk groups of breast cancer recurrence. | MATLAB | Johns Hopkins Integrated Breast Cancer Research Database | Patients: 80 |
|
| ||||||||||||
| 60 | Kaushik et al. [19], China | Chemical Biology and Drug Design | Main aim is predicting anticancer vaccine based on target sequence information using machine learning approach | More than one type of cancer | Predict vaccine response | NN | Linear and nonlinear models | Genomic | structure-based drug design | PERL programming language | Data from different resources | 100 anticancer marks |
|
| ||||||||||||
| 61 | Li et al. [93], China | 10th Annual Computing and Communication Workshop and Conference (CCWC) | The main aim of this study was predicting the response of cell lines to drugs | More than one type of cancer | Predict drug response | CNN+LSTM | Deep learning model | Genomic + Drug Response-Functional data | Drug effectiveness prediction | PaDEL+TensorFlow | GDSC + COSMIC +TCGA | 1074 cancer cell lines+ 17,419 genes in 1018 different cell lines+985 cancer cell lines |
| 62 | Laplante and Akhlouf [20], Canada | 2020 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) | The main aim was proposing a deep neural network classifier to identify the anatomical site of a tumor | More than one type of cancer | Cancer classification | NN | Deep learning model | Genomic | Discriminate between the different cancers | Not mentioned | TCGA | Not mentioned |
|
| ||||||||||||
| 63 | Liu et al. [94], USA | Genes | Discover reliable and accurate molecular network-based biomarkers for monitoring cancer treatment | More than one type of cancer | Predict drug response | NBSBM | Bayesian model + Linear and non-linear models | Genomic | Predict drug response | Not mentioned | TCGA+ GSE17705+ GDSC | 16 prostate cell lines+ 103 breast cancer patient+ 319 cancer cells |
∗Abbreviation of AI methods defined by authors: SVM: support vector machine; RF: random forest; CNN: convolutional neural network; NB: naive Bayes; AF-UCS: attribute feedback-supervised classifier system; HNB: hidden Naive Bayes; MEFS: metaensemble feature selection; DT: decision trees; fuzzy logic selection algorithm MEMBA: membership margin based-attribute selection; NMTF: nonnegative matrix trifactorization; CDSS: clinical decision support system; KNN: K-nearest neighbor; NN: neural network; NLP: natural language processing; GBGFA: gene-wise prior Bayesian group factor analysis; OFSSVM: oriented feature selection SVM; KRL: Kernelized rank learning; PLATYPUS: Progressive LAbel Training bY Predicting Unlabeled Samples; LASSO: least absolute shrinkage and selection operator Cox regression; DenseNet: densely-connected classification network; WfG: Watson for Genomics; ENR: elastic net regression; C-T CERP: Classification-Tree CERP; CART: regression trees; MVL: multiview learning; LR: logistic regression; FE: feature extraction; PEMRIC: pharmacogenomics evidence mapping for reasoning with individualized clinical data; TCIA: The Cancer Immunome Atlas; GEO: Gene Expression Omnibus; CCLE: Cancer Cell Line Encyclopedia (CCLE); NSCLC: Nonsmall cell lung cancer treatment; CCLP: COSMIC cell line project; GDSC: Genomics of Drug Sensitivity in Cancer; CPTAC: Clinical Proteomic Tumor Analysis Consortium; CGP: comprehensive genomic profiling; NCICPD: Nursing CPD Institute; KRBDSR: Kent Ridge Biomedical Data Set Repository; DTI: drug-target interaction; EHR: electronic health record; PSPEP: Proteomics Performance Evaluation Pipeline Software; TCGA: The Cancer Genome Atlas; NACT: neoadjuvant chemotherapy; GRAPE: gene-ranking analysis of pathway expression; FNABs: fine needle aspirate biopsies; PSPEP: Proteomics Performance Evaluation Pipeline Software; EGFR: epidermal growth factor receptor; EGFR-TKI: EGFR tyrosine kinase inhibitors; RWRF: reweighted random forest; RPPA: reverse-phase protein array; BLCA: bladder urothelial carcinoma; ETE: extrathyroidal extension; PTC: papillary thyroid carcinoma; BAP1: BRCA1-associated protein; TR: tumor region; NCCH: National Cancer Center Hospital; XGB: eXtreme Gradient Boosting; AML: acute myeloid leukemia; ALL: acute lymphoblastic leukemia; PAAD: pancreatic ductal adenocarcinoma; LUAD: lung adenocarcinoma; PSVM: prognostic signature vector matching; MLP: multilayer perceptron; NDM: N-desmethyltamoxifen; ECF-S: ensemble collaborative filtering method with simple averaging; ECF-W: ensemble collaborative filtering method with weighted averaging; ELNET: elastic net; RefDNN: reference drug-based neural network; AITL: adversarial inductive transfer learning; NBSBM: network-based sparse Bayesian machine.