AtropDiff: Data-Scarce Atropisomer Generation via Multi-Task Pretrained Classifier-Guided Diffusion

Abstract

AtropDiff is a diffusion framework guided by a multi-task pretrained classifier that encodes stereochemical knowledge. Dynamically weighted gradient guidance balances chirality accuracy with functional-group validity, while progressive scaffold integration supports optimization from known atropisomers.

In brief

Classifier-guided diffusion for controllable atropisomer generation in data-scarce regimes.