Back

HYALINE: Geometric Deep Learning for Accurate Prediction of G Protein-Coupled Receptor Activation States from Structure

Khaleq, A.; Kabodha, H.

2026-01-05 bioinformatics
10.64898/2026.01.05.697778 bioRxiv
Show abstract

Characterizing the conformational landscapes of G protein-coupled receptors (GPCRs) is fundamental to understanding signal transduction and accelerating rational drug design. However, current computational approaches often rely on static sequence analysis or lose critical geometric context, failing to resolve the fine-grained structural switches that drive allosteric signaling. Here we introduce Hyaline, a geometric deep learning framework that leverages E(n)-equivariant graph neural networks and ESM3 evolutionary embeddings to predict GPCR activation states directly from 3D coordinates. By incorporating biological priors through motif-specific attention biasing, Hyaline achieves near-perfect classification performance (AuROC 0.995) on a dataset of 1,590 experimental structures, significantly outperforming sequence-only models in complex cases such as Class C receptors. Hyaline provides a rapid, interpretable framework for annotating receptor conformational states, establishing a scalable foundation for the high-throughput discovery of allosteric modulators in complex signaling landscapes.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.