🧬 Bio Trust OS

AI-Powered Biological Data Trust & Readiness Platform

Bio Trust OS ingests biological tabular datasets (gene expression, proteomics, metabolomics, or clinical cohort CSVs), runs deterministic QC/anomaly/batch-effect engines, computes a Biological Trust Score (0–100), and issues a Go / Conditional / No-Go decision gate — with interactive Plotly visualizations and an exportable executive QC report.

Quick Start

pip install -r requirements.txt
streamlit run app.py

No API key or internet connection is required — the app works fully out of the box using two built-in synthetic benchmark cohorts (generated in-memory by data_generator.py), and remediation narratives are produced by an offline, deterministic rule-based agent (qc_engines.RuleBasedRemediationAgent).

Project Structure

app.py             Streamlit UI: layout, charts, export, orchestration
qc_engines.py       Deterministic QC engines: completeness, anomaly/outlier
                    (Isolation Forest + Z-score), batch effect (PCA + ANOVA +
                    silhouette), cohort balance, composite trust scoring,
                    and the offline remediation agent
data_generator.py   Synthetic "Benchmarked Clean Cohort" and
                    "Compromised / Batch-Corrupted Cohort" generators
requirements.txt    Python dependencies

Input Format

CSV/TSV with:

• Rows = samples
• Numeric columns = features (genes/proteins/metabolites)
• Optional metadata columns (auto-detected, case-insensitive aliases supported): SampleID, Batch (or Plate), Condition (or Phenotype/Label/Target/Class), Center (or Site/Lab)

Missing metadata columns degrade gracefully — each engine returns a neutral score and an explanatory note rather than failing.

Scoring

Biological Trust Score =
      0.25 * Completeness
    + 0.35 * Batch Consistency
    + 0.25 * Outlier Integrity
    + 0.15 * Balance

Trust Score	Gate
≥ 75	🟢 GO (Ready for AI/ML Training)
50–74	🟠 CONDITIONAL / NEEDS CORRECTION
< 50	🔴 NO-GO (High Risk of Bias / Failed Reproducibility)