Equitable Health Intelligence (EHI) Benchmark v1.0: Datasets, Code, and Task-Level Results for Multi-Population Omics-Based Cancer Prognosis

Equitable Health Intelligence (EHI) Benchmark v1.0 provides the datasets, computational code, and task-level experimental results underlying the EHI benchmark for studying predictive-performance disparities in omics-based cancer prognosis across population groups. The resource accompanies the Equitable Health Intelligence (EHI) Portal (https://ehiportal.org/) and supports reproducibility and secondary analysis of the benchmark results. Contents 1. EHI_Ver1_Dataset.zipContains the data resources used for benchmark construction and analysis, including genetic ancestry information, protein-expression data, mRNA expression, microRNA expression, DNA methylation, and clinical outcome information used in the EHI experiments. 2. EHI_Ver1_Codes.zipContains the computational implementation used for data preprocessing, machine-learning task construction, model training and evaluation, and generation of benchmark results. 3. EHI_Ver1_Results.zipContains task-level experimental results organized according to the five benchmark studies: Protein Expression, p-value/ANOVA-based feature selection, PCA-based feature extraction, Autoencoder (Linear–Linear–MSE), and Autoencoder (ReLU–Sigmoid–BCE). The task-level result files contain repeated experimental evaluations used for performance comparison and statistical analysis. EHI evaluates Mixture Learning, Independent Learning, Naive Transfer, and Transfer Learning conditions across omics-based cancer prognosis tasks. The benchmark investigates predictive-performance disparities associated with unequal population representation and evaluates the potential of transfer-learning approaches to improve performance for data-disadvantaged populations. Interactive resource: https://ehiportal.org/ Version: EHI Benchmark v1.0