TernaryTx combines frontier AI with deep physics to prospectively design molecular glue therapeutics, a class of drugs that work by binding two proteins together. Its four-engine platform—Leopard, Puffin, Gecko, and Octopus—covers target assessment, complex structure prediction, stability validation, and glue optimization, enabling rational design before synthesis. The platform screens up to millions of compounds to identify optimized glue candidates.
Funding
Funding not disclosed
Founders
Product
Problem
Molecular glues—drugs that work by binding two proteins together—underpin blockbuster therapies like Revlimid and Cyclosporin, yet their mechanism was only understood after these drugs reached the market. Designing such molecules prospectively has resisted rational approaches because a glue must simultaneously satisfy three structural constraints: anchoring to one protein, presenting a surface, and stabilizing a specific conformation in the partner protein. This geometric complexity has historically forced a trial-and-error discovery process.
Solution
TernaryTx provides a computational platform that breaks the three-body problem of molecular glue design into tractable components, applying machine learning as a structural-biology engine rather than a wrapper around docking. The platform's four integrated engines—Leopard, Puffin, Gecko, and Octopus—work sequentially to assess target "glueability," predict binary and ternary complex structures, validate complex stability through molecular dynamics, and optimize glue candidates. This pipeline enables researchers to rationalize and refine glue molecules before synthesis, shifting from post-hoc discovery to prospective design. The result is a platform that can identify actionable pockets, prioritize target-effector pairs, and screen millions of compounds to detect small structural changes that drive significant activity differences.
Target Audience
Primary customers are pharmaceutical and biotechnology companies engaged in targeted protein degradation and induced-proximity drug discovery, particularly those seeking to design molecular glues for previously undruggable targets.
Features
- Leopard: proprietary ML model that predicts whether a protein is "glueable" by evaluating physicochemical surface features and identifying actionable pockets capable of forming extended interfaces
- Puffin: ML tool that predicts binary and ternary complex structures for novel proteins and ranks virtual target-effector pairs to narrow the search space for target deconvolution
- Gecko: molecular dynamics-based tool that mimics atomic force microscopy in full atomistic detail to assess the physical strength of protein-protein interactions and distinguish correct from incorrect complex poses
- Octopus: validated affinity prediction and virtual screening model combining ML and physics-based methods to design optimized glues, screening up to millions of compounds
- Platform integrates ML and physics-based methods across all four engines, applying each technique where it has genuine signal rather than as a wrapper around docking