#36 · Structural bio & protein design
Geometry-aware representations for molecules, binding sites, and glycan structures
The dominant thread is encoding three-dimensional structural context into learned representations for prediction. SenCos-GEM and MolMAE build geometry into small-molecule representations, using a law-of-cosines consistency loss with SE/FiLM adapters and surface point clouds, 3D graphs and fragments respectively, with MolMAE pretrained on ~261K bioactive molecules. FUSED jointly models structural (CDPA covariance) and elemental (ilr-coordinate) descriptors of ligand binding sites. SweetFold adapts the Boltz-1x all-atom diffusion model to glycan, glycoprotein and protein-glycan complex structures. LINKER, a sequence-based model predicting residue-functional group interactions with interpretability emphasized, sits outside the geometric paradigm. What is new is the extension of geometry-aware methods to glycans and to binding-site element descriptors.
Published papers
LINKER: Learning Interactions between Functional Groups and Residues with Chemical Knowledge-Enhanced Reasoning and Explainability.
LINKER is a sequence-based model predicting residue-functional group interactions from protein sequence and ligand SMILES, trained with structure-derived…
Preprints
SenCos-GEM: SENet-Calibrated and Law-of-Cosines-Constrained Geometry-Enhanced Molecular Representation for Property Prediction
SenCos-GEM is a 3D geometry-enhanced molecular representation framework using law-of-cosines consistency loss and SE/FiLM adapters, reporting gains on…
FUSED: A Functional Representation for Joint Structural and Elemental Analysis of Protein Ligand Binding Sites
FUSED is a functional representation jointly modelling structural (CDPA covariance) and elemental (ilr-coordinate) descriptors of ligand binding sites versus…
MolMAE: A Surface-Centric Multimodal Masked Autoencoder for Molecular Representation Learning
MolMAE is a surface-guided multimodal masked autoencoder that pretrains on ~261K bioactive molecules using surface point clouds, 3D graphs, and fragment…
A Glycan-Aware Diffusion Model for Carbohydrate and Glycoprotein Structure Prediction
SweetFold adapts the Boltz-1x all-atom diffusion model to predict glycan, glycoprotein, and protein-glycan complex structures, using pseudo-polymer glycan…
Topic history
| Date | Rank |
|---|---|
| #36 | |
| #27 | |
| #33 | |
| #29 | |
| #12 | |
| #34 | |
| #19 | |
| #5 | |
| #30 | |
| #7 | |
| #20 | |
| #31 | |
| #4 | |
| #3 | |
| #7 | |
| #2 | |
| #3 | |
| #5 | |
| #9 | |
| #17 | |
| #35 | |
| #8 | |
| #23 | |
| #23 | |
| #10 | |
| #2 | |
| #10 | |
| #19 | |
| #22 | |
| #30 | |
| #10 | |
| #28 | |
| #32 | |
| #15 | |
| #8 | |
| #13 | |
| #35 | |
| #29 | |
| #36 | |
| #31 | |
| #12 | |
| #10 | |
| #24 | |
| #7 | |
| #9 | |
| #25 | |
| #9 | |
| #5 | |
| #11 | |
| #3 | |
| #14 | |
| #7 | |
| #16 | |
| #4 | |
| #5 | |
| #13 | |
| #27 | |
| #26 | |
| #34 | |
| #27 | |
| #38 | |
| #39 | |
| #37 |