Table 2 Application of machine learning technologies in immunotherapy-related tumor microenvironment analyses
From: Informing immunotherapy with multi-omics driven machine learning
Task | ML Model | Cancer | TME feature type | Input | Output | Ref |
|---|---|---|---|---|---|---|
Predict MSI status | RF-based model, SVM | Not mentioned | 808 cancer-gene panel (DNA, RNA) | 54 features based on the sequenced panel | MSI classification: MSI high vs. MSS | |
Identify gene target panel to predict TMB and response | LASSO regression | Metastatic melanoma, NSCLC | WES | Somatic mutations | Responders vs. Non-responders and selected mutations | |
Identify cancer stem-like signatures | LASSO COX regression | Gastric cancer | RNA | RNA of 2,527 genes | Responders vs. Non-responders and selected stem-like features | |
Identify cancer stem-like signatures | Cancer stemness clustering: K-means; Cancer stemness feature selection: LASSO regression, SVM, RFB, XGBoost, LR; Response prediction: TIDE | GBM | RNA | RNA of cancer stemness-associated DEGs | Stemness subtype cluster and selected cancer stemness-associated genes | |
Identify CAF signatures | CAF subtype clustering: Consensus clustering; Gene selection: RF, DT, KNN | Melanoma, lung cancer, TNBC | RNA | Prognostic-related RNA data | CAF-subtype clustering and selected subtype-related genes | |
Identify CAF signatures | LASSO regression, RF | Melanoma | RNA | DEGs | Responders vs. Non-responders and key CAFs-related DEGs | |
Identify gene signatures and immunotherapy response prediction | TME clustering: Hierarchical clustering; Cluster feature selection: LASSO Cox regression, RF; TME cluster classification: SVM, NB, RF, NN; Risk prediction: DT | LUAD | RNA | RNA, clinicopathological traits | TME (risk) cluster classification: low vs. intermediate vs. high and their cluster related gene features | |
Identify immune-related genes from protein signatures | Immune-related gene identification: NN Immunotherapy response: DT | Gastric cancer | PPI network data, RNA | PPMI matrix based on PPI network data, RNA | NN: Gene property classification (immune-promoted vs. immune-inhibited); DT: Response prediction (Responders vs. Non-responders) | |
Identify TIIClnc | LASSO regularized LR, Boruta, XGBoost, SVM, RF | GBM | RNA | Selected lncRNA | Regulation prediction in immune cell lines and GBM cell lines (upregulated vs. downregulated) | |
Identify TIIClnc | LASSO, Ridge, stepwise Cox, CoxBoost, RSF, Enet, plsRcox SuperPC, GBR, survival-SVM | LGG | RNA | Filtered top expressed TIIClnc signatures | Responders vs. Non-responders and selected TIIClnc signatures | |
Identify impact of CTLA-4 blockade on antigen-specific, human T-cell responses early between neonates and adults | RF | Healthy donors | Flow cytometry | Frequencies of cytokine producers in the encountered CD4 + T-cell responses | CD4 + T cell classification (neonates vs. adults) after CTLA-4 blockade stimulation | |
Predict T cell infiltration | LR, SVM | Colorectal cancer | Histological data, 373 cancer and immune related gene panel from FoundationOne | LR: image-based features SVM: patient’s gene expression profile | T cells and tumor cells co-localized vs. not co-localized | |
Predict TIL | Multimodal NN model | Colorectal cancer, breast cancer, lung cancer, pancreatic cancer | RNA, H&E staining images | RNA-seq + Visual texture feature extracted from H&E staining | Proportions of five immune cell types within tumors and total TIL proportions | |
Identify epigenomic signatures | RF | LUAD | DNA methylation data | iDMCs | Immunoactivity classification and selected signatures | |
TIME deconvolution | nu-SVR-based noise constrained recursive feature selection | Not mentioned | RNA | RNA | Proportions of 22 immune cell types | |
Identify tumor-associated metabolism subtypes | Cox regression with LASSO penalty | LUAD | RNA | RNA of 1,426 lipid metabolism genes and 1,638 immune-related genes | Metabolic TME subtype prediction (metabolism vs. immunoactive) |
