search

choderalab / fah-xchem

Tools and infrastructure for automated compound discovery using Folding@home
MIT License
6 stars 3 forks source link

Fix compound free energy estimates from diffnet #102

Closed jchodera closed 3 years ago

jchodera commented 3 years ago

While the transformation free energies appear to be correctly computed, the free energies from diffnet are (1) way too negative, and (2) completely inconsistent with the transformation free energies.

For example, this molecule has a predicted affinity of 15 fM (-19.1 kcal/mol): image However, all the transformations indicate this compound should bind more weakly than the reference with affinity 1.7 µM (-8.0 kcal/mol): image

jchodera commented 3 years ago

Updating to arsenic 0.1.0 in the production environment seems to have fixed this!