BioFlow ML Studio v2
AI-driven biomedical decision platform
Data integration • QC • predictive modeling • representation learning • in silico simulation • AI copilot
BioFlow ML Studio v2 extends the original biomarker-discovery MVP into a more complete translational AI workflow.
It keeps the practical V1 pipeline for upload, QC, feature selection, signatures, calibration, validation, DCA, and reports,
while adding a V2 representation layer, simulation lab, and AI copilot for hypothesis generation and publication-ready interpretation drafts.
What changed in v2
Embedding Space
Simulation Lab
AI Copilot
Project Pack
External Validation
Representation Layer- Builds a PCA-based latent biomedical embedding
- Shows sample-level structure and label separation
- Ranks component loading features
In Silico Simulation- Simulates feature upregulation, downregulation, or knockout
- Compares prediction scores before and after intervention
- Supports hypothesis generation for candidate biomarkers
AI Copilot- Summarizes model metrics and candidate features
- Integrates embedding and simulation context
- Generates cautious interpretation drafts without overclaiming causality
Core platform modules
Data Upload & QC- Upload expression matrices and metadata
- Validate sample IDs and labels
- Run PCA-based quality control
Feature Selection & Signatures- Rank candidate features
- Build sparse signatures
- Inspect coefficient-based biomarkers
Modeling & Validation- Train predictive models
- Run external validation
- Review calibration and decision curves
Reports & Reproducibility- Export HTML reports
- Save project snapshots
- Package runs as project packs
Visual preview
Representative outputs from the platform, including model validation, biomarker heatmap visualization, and exportable reports.

Modeling & Validation
Compare predictive models, inspect performance metrics, and review validation outputs.

Heatmap & Signature View
Visualize selected biomarker features and sample-level expression patterns.

Automated Report Export
Export structured summaries and presentation-ready analysis reports.
Recommended workflow
- Load demo data or upload expression and metadata tables
- Validate and build the ML-ready dataset
- Run QC and feature selection
- Train predictive models
- Inspect signatures, heatmaps, calibration, DCA, and external validation
- Run Embedding Space to inspect latent structure
- Run Simulation Lab to test candidate feature perturbations
- Use AI Copilot to draft biological interpretation and next-step hypotheses
- Export reports, snapshots, or project packs
Scientific positioning
The V2 modules are designed for hypothesis generation, not direct causal proof. BioFlow can help prioritize candidate genes,
summarize predictive signals, and structure reproducible analysis, but biological mechanisms still require literature support,
external validation, and experimental confirmation.