Biomarker Signature Studio
Design validated biomarker panels that are explainable, stable, and ready for translational follow-up. This skill stitches together the existing biomarker pipeline tooling, adds configurable feature-selection ensembles, a small survival-analysis hook, and artifact export so downstream lab teams can review QC outputs.
What This Skill Does
- QC + Harmonization: Align expression matrices (samples x features) with metadata, check label balance, and compute summary stats.
- Feature Selection Ensemble: Supports Boruta, elastic-net stability, mutual-information top-K, and mRMR with optional intersection voting.
- Model Factory: Trains multiple estimators (Logistic L1, RandomForest, XGBoost if present) under nested CV, picks champion by AUC.
- Explainability + Export: Produces SHAP tables/plots when packages are available, exports feature rankings and model weights.
- Survival Hook: If metadata contains
time_to_event and event the skill computes concordance for selected features via Cox model.
All logic lives in scripts/biomarker_signature_studio.py.
Inputs
- Expression matrix (
--expression): CSV/TSV genes x samples or samples x genes (auto-detected by metadata match).
- Metadata (
--metadata): Must contain --label-column. Optional --id-column (default sample_id), time_to_event, event.
- Optional gene list for filtering (
--feature-list).
- Output directory (
--output-dir), created if missing.
Quick CLI Usage
python Skills/Research_Tools/Biomarker_Signature_Studio/scripts/biomarker_signature_studio.py \
--expression data/expression.csv \
--metadata data/metadata.csv \
--label-column phenotype \
--selectors boruta,lasso,mrmr \
--models rf,logit \
--output-dir outputs/biomarkers_run1
Key flags:
| Flag |
Description |
--selectors |
Comma list of selection strategies (boruta, lasso, mrmr, mi_topk). |
--models |
Models to evaluate (logit, rf, xgb). |
--k-features |
Target number of features for mrmr/mi_topk. |
--survival |
Enable Cox evaluation when survival columns exist. |
--random-state |
Reproducibility. |
--nested-folds |
Outer CV folds (default 5). |
Workflow
- Load + align inputs, infer orientation, impute missing values.
- Standardize features (fit on train set only).
- Run requested selectors; create intersection + union candidate lists.
- For each selector output run nested CV training across requested models.
- Export champion metrics (
metrics.json), feature table (selected_features.csv), SHAP summary (shap_summary.csv when available), and survival stats (survival.json).
QC Expectations
- Class count ratio ≤3:1; warnings logged otherwise.
- Selected features between 5 and 250 unless user overrides.
- Nested CV AUC ≥0.70 or flagged in report.
- SHAP overlap with selected features ≥60% (reported).
Related Assets
examples/configs/biomarker_studio_template.yaml (scaffold for teams)scripts/biomarker_signature_studio.py (entry point)- Existing biomarker workflow skill for orchestrated runs.
Use this skill whenever you need a ready-to-review biomarker dossier (data QC, model metrics, explainability artifacts) before moving to validation cohorts or lab assays.
1---2name: bio-research-tools-biomarker-signature-studio3description: Multi-omic biomarker discovery studio that ingests expression + metadata, performs QC, multi-strategy feature selection, nested CV model training, survival analysis hooks, and SHAP-based interpretation. Use to design translational biomarker panels with documented evidence.4---56<!--7# COPYRIGHT NOTICE8# This file is part of the "Universal Biomedical Skills" project.9# Copyright (c) 2026 MD BABU MIA, PhD <md.babu.mia@mssm.edu>10# All Rights Reserved.11#12# This code is proprietary and confidential.13# Unauthorized copying of this file, via any medium is strictly prohibited.14#15# Provenance: Authenticated by MD BABU MIA1617-->181920# Biomarker Signature Studio2122Design validated biomarker panels that are explainable, stable, and ready for translational follow-up. This skill stitches together the existing biomarker pipeline tooling, adds configurable feature-selection ensembles, a small survival-analysis hook, and artifact export so downstream lab teams can review QC outputs.2324## What This Skill Does25261. **QC + Harmonization:** Align expression matrices (samples x features) with metadata, check label balance, and compute summary stats.272. **Feature Selection Ensemble:** Supports Boruta, elastic-net stability, mutual-information top-K, and mRMR with optional intersection voting.283. **Model Factory:** Trains multiple estimators (Logistic L1, RandomForest, XGBoost if present) under nested CV, picks champion by AUC.294. **Explainability + Export:** Produces SHAP tables/plots when packages are available, exports feature rankings and model weights.305. **Survival Hook:** If metadata contains `time_to_event` and `event` the skill computes concordance for selected features via Cox model.3132All logic lives in `scripts/biomarker_signature_studio.py`.3334## Inputs3536- Expression matrix (`--expression`): CSV/TSV genes x samples or samples x genes (auto-detected by metadata match).37- Metadata (`--metadata`): Must contain `--label-column`. Optional `--id-column` (default `sample_id`), `time_to_event`, `event`.38- Optional gene list for filtering (`--feature-list`).39- Output directory (`--output-dir`), created if missing.4041## Quick CLI Usage4243```bash44python Skills/Research_Tools/Biomarker_Signature_Studio/scripts/biomarker_signature_studio.py \45 --expression data/expression.csv \46 --metadata data/metadata.csv \47 --label-column phenotype \48 --selectors boruta,lasso,mrmr \49 --models rf,logit \50 --output-dir outputs/biomarkers_run151```5253Key flags:5455| Flag | Description |56|------|-------------|57| `--selectors` | Comma list of selection strategies (`boruta`, `lasso`, `mrmr`, `mi_topk`). |58| `--models` | Models to evaluate (`logit`, `rf`, `xgb`). |59| `--k-features` | Target number of features for `mrmr`/`mi_topk`. |60| `--survival` | Enable Cox evaluation when survival columns exist. |61| `--random-state` | Reproducibility. |62| `--nested-folds` | Outer CV folds (default 5). |6364## Workflow65661. Load + align inputs, infer orientation, impute missing values.672. Standardize features (fit on train set only).683. Run requested selectors; create intersection + union candidate lists.694. For each selector output run nested CV training across requested models.705. Export champion metrics (`metrics.json`), feature table (`selected_features.csv`), SHAP summary (`shap_summary.csv` when available), and survival stats (`survival.json`).7172## QC Expectations7374- Class count ratio ≤3:1; warnings logged otherwise.75- Selected features between 5 and 250 unless user overrides.76- Nested CV AUC ≥0.70 or flagged in report.77- SHAP overlap with selected features ≥60% (reported).7879## Related Assets8081- `examples/configs/biomarker_studio_template.yaml` (scaffold for teams)82- `scripts/biomarker_signature_studio.py` (entry point)83- Existing biomarker workflow skill for orchestrated runs.8485Use this skill whenever you need a ready-to-review biomarker dossier (data QC, model metrics, explainability artifacts) before moving to validation cohorts or lab assays.868788<!-- AUTHOR_SIGNATURE: 9a7f3c2e-MD-BABU-MIA-2026-MSSM-SECURE -->
FridrichMethod