Conventional scoring systems fail to take into account the complexity of these molecules designed for targeted protein degradation. RetroTAC draws on real-world synthetic pathways and clearly distinguishes the good candidates from the bad.
PROTACs are molecules that can degrade a protein targeted by the cell, a therapeutic approach that goes beyond simple inhibition. But their large size and modular structure, made of a fragment that binds to the target, a connecting arm and a ligase ligand E3, make their synthesis delicate. The usual synthetic accessibility scores tend to overestimate this difficulty by penalising complexity, while the retrosynthesis tools that offer explicit pathways are very costly in calculating for such large molecules.
The authors propose a readable score, based on the synthetic pathways, which favours a small number of steps and a balanced assembly of fragments, as well as RetroTAC, a set of machine learning models that predicts this score directly from the structure (average quadratic error of 0.13, R2 of 0.65 on test molecules). On PROTACs generated in recombinant fragments, well-rated candidates obtained an average synthesizability of 0.73 compared to 0.31 for the least-rated, and 98.75 per cent of them had their synthesis route fully resolved, compared to 51 per cent. The tool would be used to filter virtual chemotheques before reserving detailed planning for the best candidates.
Read the full article on the original source
Go to the source →