Inverse folding - BioML ZK

Antibody-antigen binding

Conformational entropy could still matter in miniprotein binder design

Antibody V-regions improve their affinity for targets by both creating more high-energy interactions and reducing the conformational entropy of their antigen-binding loops[1][2]. Entropy's importance in antibody-antigen affinity seems obvious, given that loop residues largely mediate binding[3][4]. But for de novo-designed miniprotein binders, which often

Inverse folding

Glutamate- and lysine-rich designs are susceptible to expression failure resulting from adenosine-rich sequences

High rates of glutamate and lysine introduction are a staple of structure-based sequence design by ProteinMPNN and related models, regardless of who trains them[1]. Recently, an analysis of the Bits in Bio competition showed that high glutamate/lysine content is predictive of expression failure[2]: The authors traced this

Inverse folding

Training inverse folding models with label smoothing improves fitness prediction performance

BERT protein language models and inverse folding models learn to predict masked tokens[1]; coevolutionary patterns and propensities are expected to emerge from training in an unsupervised manner[2]. This works fine when training on sequence data, which is plentiful. However, it is not clear that there are enough experimental

Inverse folding

Training protein structure-based neural networks exclusively on predicted protein structures worsens performance on experimental structures due to the training data's idealized local geometry

Predicted protein structures, particularly monomeric structures, have become ubiquitous thanks to the release of the AlphaFold Database[1] and its successors[2]. Yet training structure-based neural networks exclusively on these synthetic structures has now been widely shown to worsen performance on experimental structures. Hsu et al., who trained the structure-based