About this session
A webinar with Jingyi Chen, Scientific Developer II
Every lead optimization team faces the same question: which molecules should we make next? Even a carefully designed focused library can easily contain thousands of candidates, but conventional relative binding free energy (RBFE) calculations at that scale are cost- and time-prohibitive.
FE-NES takes a different approach. It uses endpoint sampling and many short A⇌B switches to deliver rank ordering that is competitive with methods like FEP+ and OpenFE, but with much more efficient use of compute resource. In one real example over six hundred molecules were evaluated in about three hours, making same-day prioritization possible.
To search at scales beyond hundreds FE-NES is used to provide training data for Spatial QSAR, a fast affinity prediction model utilizing our best-in-class shape and chemical feature molecular representation. In an active learning loop FE-NES results are used to train a Spatial QSAR mode on a fraction of a library of interest, Spatial QSAR predicts the rest of the library, and the most promising molecules go back to FE-NES for the next round.
Jingyi presents a BACE-1 case study on a library of over five thousand analogs, which is a very substantial compute burden even for FE-NES. Using only five hundred FE-NES evaluations over five iterations, the workflow identified close to one hundred molecules with relevant affinities from the parent library at a low cost and in less than one day. Jingyi will also show how this workflow provides a practical stopping signal for active learning.