ReduXis: A Comprehensive Framework for Robust Event-Based Modeling and Profiling of High-Dimensional Biomedical Data

Event-based models (EBMs) are powerful tools for inferring probabilistic sequences of monotonic biomarker changes in progressive diseases, but their use is often hindered by data quality issues, high dimensionality, and limited interpretability. We introduce ReduXis, a streamlined pipeline that overcomes these challenges via three key innovations. First, upon dataset upload, ReduXis performs an automated data readiness assessment-verifying file formats, metadata completeness, column consistency, and measurement compatibility-while flagging preprocessing errors, such as improper scaling, and offering actionable feedback. Second, to prevent overfitting in high-dimensional spaces, ReduXis implements an ensemble voting-based feature selection strategy, combining gradient boosting, logistic regression, and random forest classifiers to identify a robust subset of biomarkers. Third, the pipeline generates interpretable outputs-subject-level staging and subtype assignments, comparative biomarker profiles across disease stages, and classification performance visualizations-facilitating transparency and downstream analysis. We validate ReduXis on three diverse cohorts: the Emory Healthy Brain Study (EHBS) cohort of patients with Alzheimer's disease (AD), a Genomic Data Commons (GDC) cohort of transitional cell carcinoma (TCC) patients, and a GDC cohort of colorectal adenocarcinoma (CRAC) patients.