Fig. 1
From: Large-scale public data reuse to model immunotherapy response and resistance

TIDE web platform architecture. The TIDE web platform aims to facilitate the hypothesis generation, biomarker optimization, and patient stratification in immune-oncology research through a public data reuse approach. The platform functions are based on the integration of large-scale omics data and biomarkers on published ICB trials, non-immunotherapy tumor profiles, and CRISPR screens. The web platform takes gene set or expression profiles as input and provides three interactive modules. A Gene prioritization for a user-input gene set. Every gene is ranked by their clinical relevance and CRISPR screen phenotype, including four types of metrics: 1, the association between gene expression and T cell dysfunction across cohorts, computed as the z-score in the Cox Proportional Hazard (PH) regression model; 2, the association between gene expression and ICB response across tumors, computed as the z-score in the Cox-PH regression; 3, the log-fold change in CRISPR screens probing the effect of gene knockout on lymphocyte-mediated tumor killing; 4, the gene expression in cell types driving T cell exclusion in tumors. Data cohorts are grouped by their metric types on the heatmap (columns). Genes (rows) can be interactively reordered by the gene values either on a single data set or any metric type groups. B Biomarker evaluation for a custom biomarker gene set. The predictive power of biomarkers in the public immunotherapy cohorts is quantified by two criteria, the area under the receiver operating characteristic curve (AUC) and the z-score in the Cox-PH regression. We visualize biomarkers’ AUC by bar plots (left panel) and Cox-PH z-scores by Kaplan-Meier curve (right panel). C Biomarker consensus to predict ICB response from gene expression profile. Every input transcriptomic profile is evaluated by TIDE, microsatellite instability (MSI) signature, interferon-gamma (IFNG) signature, and other biomarkers reported in the literature