Intercept is the interpretability and explainability platform tailored for bio-AI and clinical-AI models

01

Why Intercept?

01

Derisk Downstream Deployment

  • Make better wet-lab decisions by understanding why your bio-AI model prioritized a drug target, molecule, or candidate
  • Help clinicians trust clinical-AI model predictions in patient-care decisions
02

Smarter Model Development

  • Inspect how biological concepts and reasoning evolve during training
  • Compare architectures, debug failures, and catch misleading patterns
  • Improve models via interpretable feedback loops — not just performance metrics
03

Scientific Discovery

  • Bio-AI and clinical-AI models often outperform humans. Interpreting them reveals new scientific insights hidden inside your models.
02

Research

Published — Proceedings of the National Academy of Sciences
Sparse autoencoders uncover biologically interpretable features in protein language model representations

Applies sparse autoencoders and transcoders to interpret protein language models at the protein-level and amino acid-level.

In Preparation
Model and Layer Diffing Protein Language Models Using Crosscoders

Uses crosscoders to reveal how biological concepts evolve across different layers and models for protein language models.

03

In the Media

MIT News
Researchers glimpse the inner workings of protein language models
→
IBM Think
AI starts to explain itself in drug discovery labs
→
The Scientist
Researchers Decode How Protein Language Models Think, Making AI More Transparent
→
04

Team

Onkar Gujral

Onkar Gujral

5th year PhD student in the Berger Lab at MIT (advised by Bonnie Berger) where he has authored foundational work on bio-AI interpretability.

05

Get in Touch

Contact Us