Skip to content
← Projects

Interpretable Regulatory Genomics

Interpretable ML to discover regulatory switches in non-coding DNA

Domain
Bioinformatics
Status
Active
Year
2026
Stars
★ 12

This project applies interpretable machine learning to ENCODE candidate cis-regulatory elements to find sequence patterns distinguishing active regulatory regions from other non-coding DNA. It progresses from classical baselines to CNNs and foundation-model embeddings, all on Kaggle T4-level compute.

The problem

Identifying which sequence features make non-coding DNA regions regulatory, in a way that is interpretable rather than black-box.

How it works

  1. 01

    Phase 1: Logistic Regression, Random Forest and XGBoost on k-mer features with SHAP

  2. 02

    Phase 2: CNNs with filter analysis, saliency maps and integrated gradients

  3. 03

    Phase 2+: 3-class classification and K562 to GM12878 zero-shot evaluation

  4. 04

    Phase 3: Nucleotide Transformer (500M) embeddings

Results

As reported in the repository

0.8830XGBoost (k=4 k-mers) test AUROC
0.8604AttentionCNN test AUROC
81.25% accAttentionCNN 3-class
0.9176Nucleotide Transformer (500M) test AUROC

Figures

Interpretable Regulatory Genomics figureInterpretable Regulatory Genomics figureInterpretable Regulatory Genomics figure