Molecular glues,
engineered from the surface up.

Frontier AI × deep physics to design small molecule therapeutics that unlock all biology.

/ Platform

Four engines. Proprietary data.

Target Glueability Assessment

Leopard

Predicts whether a protein of interest is "glueable" independently of knowing the effector protein.

Leopard is a target assessment tool that predicts whether a protein of interest is "glueable" independently of knowing the effector protein. It uses a proprietary machine learning model to evaluate the physicochemical features of a protein's surface, identifying actionable pockets that have the potential to form extended interfaces upon protein-protein interaction.

Effector Protein Selection

Puffin

Accurately predicts binary and ternary complex structures for novel proteins.

Puffin is a machine learning tool that accurately predicts binary and ternary complex structures for novel proteins. It ranks and prioritises virtual target-effector protein pairs, which rapidly narrows down the search space for target deconvolution when the required effector protein is unknown.

Ternary Complex Structure Prediction

Gecko

Assesses the stability of protein-protein complexes using molecular dynamics.

Gecko is a proprietary, molecular dynamics-based tool that assesses the stability of the protein-protein complexes generated by Puffin. By mimicking atomic force microscopy in full atomistic detail, Gecko evaluates the physical strength of the interactions to accurately distinguish between correct and incorrect complex poses.

Glue Optimisation

Octopus

Combines machine learning and physics-based methods to design optimised molecular glues.

Octopus is a validated affinity prediction and virtual screening model that combines machine learning and physics-based methods to design optimised molecular glues. Operating on the complex structures built by Puffin and Gecko, it rapidly screens up to millions of compounds to identify hits and detect small structural changes that lead to significant differences in activity.

/ Approach

Designed, not discovered.

The past was post-rationalisation.
The future is prospective design.

The blockbuster drugs Revlimid, Sirolimus and Cyclosporin are prescribed to millions of patients around the world, helping to treat many diseases, from Cancer to Arthritis. But it's only recently been discovered how they actually work: by glueing two proteins together. The reason this has resisted prospective design until now is geometric: a glue molecule has to satisfy three structural constraints at once — its anchor on protein A, its presented surface, and the conformation it stabilises in protein B.

Our platform breaks that three-body problem into tractable pieces, applying ML where ML actually has signal — as a structural-biology engine in its own right, not as a wrapper around docking. The result is glues we can rationalise before we make them, and refine without starting over.

Chris Tame & Andrew Potterton

Co-Founders · CEO & CTO

/ News