Predictive Models
from Omics Data

Build interpretable, parsimonious classification models from high-dimensional omics data. Discover minimal-feature signatures using evolutionary algorithms.

1000x
Faster (Rust engine)
<1%
Feature selection
3
Languages (R, Python, Web)
Open
Source & Free

Designed for Discovery

From raw omics matrices to publication-ready models in minutes, with full transparency at every step.

Parsimonious Models

Discovers minimal feature sets (binary, ternary, ratio) that achieve high classification accuracy. Models you can actually interpret.

🧬

Evolutionary Search

Genetic algorithms and beam search heuristics explore the feature space efficiently to find optimal signatures.

🗳

Jury Voting

Ensemble of expert models with majority or consensus voting, rejection capability, and per-sample confidence scores.

📊

Rich Visualizations

Prevalence plots, heatmaps, violin plots, AUC evolution, model complexity tracking, and sample predictions.

🔄

Cross-validation

Built-in k-fold cross-validation with generation-level tracking. Monitor train vs test performance in real time.

Blazing Fast

Core engine rewritten in Rust, delivering up to 1,000x speedup over the original R implementation. Analyze thousands of features in seconds.

Web Application

A full-stack interface to upload data, configure experiments, launch analyses, and explore results interactively.

Data explorer with prevalence plots, feature filtering and barcode visualization
Parameter configuration for genetic algorithm, voting ensemble, and feature importance
Job table with AUC scores, model complexity, and summary metrics
AUC evolution over generations, model complexity tracking, fit vs AUC
Jury voting with confusion matrices, vote matrix heatmap, sample predictions and FBM population
Population of models with feature heatmap and violin plots

The Predomics Suite

Three complementary tools, one unified approach to interpretable omics classification.

Rust + Python

gpredomics

High-performance ML engine rewritten from scratch in Rust — up to 1,000x faster than the original R package. Exposes Python bindings via gpredomicspy for seamless integration with notebooks and scripts.

View Repository →
FastAPI + Vue.js

PredomicsApp

Full-stack web application for running analyses, exploring results, managing projects, and sharing with collaborators. Deploy anywhere with Docker.

View Repository →
R Package

gpredomicsR

R bindings for the gpredomics Rust engine. Successor to the original predomicspkg, bringing the 1,000x Rust speedup to the R ecosystem with a familiar interface.

View Repository →

Up and Running in Minutes

Choose your preferred way to use Predomics.

🐳 Docker (Web App)

git clone https://github.com/predomics/predomicsapp-web.git
cd predomicsapp-web
docker compose up -d

# Open http://localhost:8001

🐍 Python

import gpredomicspy

param = gpredomicspy.Param()
param.load("params.yaml")
experiment = gpredomicspy.fit(param)
experiment.display_results()

📊 R (gpredomicsR)

# Coming soon — R bindings for gpredomics
devtools::install_github("predomics/gpredomicsR")

library(gpredomicsR)
result <- gpredomics.fit(params)

Publication

GigaScience, 2020
Interpretable and accurate prediction scores for metagenomics data using a ternary encoding approach
Prifti E., Chevaleyre Y., Hanczar B., Zucker J.D. et al.