Who Actually Responds? Multi-Omics Patient Stratification on BioMate

Most drugs work for some patients and not others. Finding the "some" before the trial is patient stratification — and it is the single highest-leverage move in modern drug development. Our Patient Stratification survey maps the whole field, from PAM50 and consensus molecular subtypes to causal machine learning and adaptive trials. Here's how BioMate puts it to work.

The one distinction that matters

The survey's throughline is predictive vs. prognostic: a prognostic marker tells you the likely outcome regardless of treatment; a predictive marker tells you who benefits from a specific therapy — and only the latter licenses a treatment decision. Plain outcome-prediction models capture the prognostic signal; it takes multi-omics factor models, causal methods, and enrichment designs to isolate the predictive one. The mature, adopted layer is genomic biomarkers with FDA companion diagnostics (MSI-H, HER2, the 18-gene T-cell-inflamed GEP) plus validated multigene assays.

What BioMate does

• Cross-modal responder stratification. BioMate runs MOFA2 joint decomposition across RNA-seq + Olink proteomics on the same patients. In a worked CIDP analysis, Factor 1 (complement) separated responders at AUROC 0.79 and Factor 2 (FcRn) at 0.71 — and the pipeline reproduced the canonical CLL benchmark (IGHV status, AUROC 0.98) on the way, so you can trust the harness before trusting the result.
• Single-cell responder biology. scRNA-seq clustering → responder differential expression → pathway enrichment runs in one session (e.g., an anti-TNF non-responder signature with CXCL13 +4.2 log₂FC).
• Molecular subtyping. Transcriptomic subtype classifiers and interferon-signature scoring connect a cohort to the treatment-relevant subgroup — the same logic behind consensus molecular subtypes and IFN-high trial enrichment.
• Reproducible scoring. Every stratifier is scored against gold-standard labels with a documented metric harness — the survey's core caution is that most published models fail on leakage and missing validation, so BioMate bakes the validation in.

Where BioMate fits in the comparison

The survey shows stratification winning inside adaptive master protocols with multi-omics, causally-grounded signatures. BioMate operationalizes the multi-omics half — a plain-English path from raw RNA-seq + proteomics to a scored, cross-validated responder signature — so a translational team can build and stress-test a stratifier on their cohort in a session rather than a quarter.

Honest boundaries

These are research/translational analyses, not FDA-cleared companion diagnostics; predictive claims require a treatment-by-biomarker interaction and prospective validation, which BioMate helps you design and test — not skip.