Our discovery engine is not a single algorithm. It is structural modelling, machine learning and bench validation kept on the same clock.
We have synthesised and encoded a library of 343 million compounds. A single affinity selection puts all of them in front of a target at once, and the binders are read out by sequencing — chemical space that plate-based screening could not sample in years.
Selection data from a library this size is dominated by noise. Machine learning and deep learning models trained on our own data separate genuine binders from artefacts, so chemistry effort goes only into compounds worth making.
Our physical chemists build and refine structural models of the target, giving every downstream decision a physical basis rather than a purely statistical one.
Docking predictions explain how a hit binds — which contacts matter, which parts of the molecule are free to change, and where affinity or selectivity can be won.
Hits are modified round by round against that structural picture. Each round is designed in silico and confirmed at the bench before the next begins.
Optimised binders are developed as peptide-drug conjugates or as PROTAC and degrader molecules, depending on what the target allows.
Selection, modelling and synthesis are not handed between departments. They are one cycle, run by one team.
343 million encoded compounds meet the target in a single experiment; sequencing reports what stuck.
Models trained on our own selection data rank candidate binders and discard the artefacts a library this size inevitably produces.
Physical chemists model the target and dock the hit to work out which contacts drive binding and which positions are free to change.
The hit is redesigned against that structural picture, synthesised, and measured — then the cycle runs again.
We are happy to walk partners through the platform under CDA, including benchmark data and cycle-time metrics.