ConforNets: Latents-Based Conformational Control in OpenFold3
2026-04-20 • Machine Learning
Machine Learning
AI summaryⓘ
The authors studied how to better get AlphaFold 3 (AF3) to predict different shapes a protein might take, not just the most common one. They found that changing certain internal parts called 'latent representations' in a specific way (called ConforNets) helps the model show more useful variations. This method works well across many proteins and can even apply changes learned from one protein to related ones. Their approach improves how AF3 can control and predict alternate protein shapes.
AlphaFoldprotein conformationlatent representationPairformerprotein foldingconformational variabilityunsupervised learningsupervised learningmulti-state benchmarkprotein family
Authors
Minji Lee, Colin Kalicki, Minkyu Jeon, Aymen Qabel, Alisia Fadini, Mohammed AlQuraishi
Abstract
Models from the AlphaFold (AF) family reliably predict one dominant conformation for most well-ordered proteins but struggle to capture biologically relevant alternate states. Several efforts have focused on eliciting greater conformational variability through ad hoc inference-time perturbations of AF models or their inputs. Despite their progress, these approaches remain inefficient and fail to consistently recover major conformational modes. Here, we investigate both the optimal location and manner-of-operation for perturbing latent representations in the AF3 architecture. We distill our findings in ConforNets: channel-wise affine transforms of the pre-Pairformer pair latents. Unlike previous methods, ConforNets globally modulate AF3 representations, making them reusable across proteins. On unsupervised generation of alternate states, ConforNets achieve state-of-the-art success rates on all existing multi-state benchmarks. On the novel supervised task of conformational transfer, ConforNets trained on one source protein can induce a conserved conformational change across a protein family. Collectively, these results introduce a mechanism for conformational control in AF3-based models.