Documentation
Welcome to the Predomics documentation. Navigate through the sidebar to learn about the project.
Overview
Predomics discovers parsimonious predictive models from omics data. Given a high-dimensional feature matrix (e.g. metagenomic species abundances) and binary class labels (e.g. healthy vs. disease), the algorithms search for the smallest set of features that accurately discriminate between classes.
The models use simple mathematical languages:
- Binary (0/1) – feature presence/absence
- Ternary (-1/0/+1) – feature contributes negatively, is absent, or contributes positively
- Ratio – ratio of two feature groups
These compact signatures are easy to interpret, validate, and translate into clinical biomarkers.
Architecture
predomicspkg (R) -- Original algorithms and visualizations
gpredomics (Rust) -- High-performance engine
gpredomicspy (Python) -- Python bindings via PyO3/Maturin
predomicsapp-web -- Web application
backend (FastAPI) -- REST API, job runner, PostgreSQL
frontend (Vue.js 3) -- Interactive UI with Plotly charts