Blog
-
Topos-Chem and Topos-AP: designing past the edge of known chemistry
One of the biggest challenges in drug discovery is finding molecules that work without simply making small variations of drugs we already know. The chemical possibilities are almost limitless, but scientists can only make and test a tiny fraction of them. AI offers a way to explore this vast, largely uncharted territory. However, predicting whether an unfamiliar molecule will actually work is much harder than predicting the behavior of something similar to an existing drug. This is the problem we're tackling with Topos-Chem and Topos-AP. In a recent kinase drug discovery program, a medicinal chemist working with our platform designed the two most potent new compounds in the program, both substantially different from published chemistry. The results suggest a path toward discovering effective medicines in regions of chemical space that have been difficult to explore.
Read Post → -
Why kinase binding is dynamic
Some of the most successful modern medicines work by blocking protein kinases, molecular switches that control how cells grow, divide, and respond to signals. These proteins are important drug targets in cancer and many other diseases. But designing drugs against them is more complicated than it might seem. Proteins aren't rigid structures; they're constantly moving, shifting between different shapes that can dramatically change how a drug interacts with them. And even when two drugs appear equally potent in a laboratory test, they may bind at different speeds, remain attached for different lengths of time, or inhibit the protein through entirely different mechanisms. Understanding these differences is critical to designing better drugs. It also exposes a fundamental limitation of many current AI approaches to drug discovery. A single static picture of a protein, or a single predicted binding affinity, cannot capture the thermodynamics and kinetics that determine how drugs actually work.
Read Post → -
Topos-2 and Topos-Bind
Today we are announcing Topos-2, our next-generation model for generating conformational ensembles across the order-disorder spectrum. We are also previewing Topos-Bind: a generative model for conformational ensembles of disordered proteins bound to small molecules.
Read Post →