Antibody structure prediction

The antibody structure prediction problem is a subclass of the protein structure prediction problem that is mostly focused on predicting the CDRs, specifically the CDRH3 loop that mediates antigen binding. It also sometimes includes the antibody-antigen docking problem.

Models

AbFold: a method that concatenates information from IgFold and AlphaFold2 Multimer when modeling antibodies, which usually predict different conformations, leading to improved prediction quality (1)
IgFold: a method that uses embeddings from the antibody-specific PLM AntiBERTy
ImmuneBuilder: Consists of ABodyBuilder2 (for Antibodies), NanoBodyBuilder2 (for Nanobodies), and TCRBuilder2 (for T-cell receptors).

Peng C, Wang Z, Zhao P, Ge W, Huang C. AbFold – an AlphaFold Based Transfer Learning Model for Accurate Antibody Structure Prediction. openRxiv; 2023. Available from: https://doi.org/10.1101/2023.04.20.537598

Quartz 4

Explorer

Antibody structure prediction

Models

General

CDR representations segregate into distinct clusters

Cdr

Antibody structure prediction improved with AlphaFold2 features

LM-based antibody structure prediction methods work equally well with generic and Ab PLMs

CDRH3 conformation is affected by interdomain orientation

Different CDRL3 lengths are found in functional and dysfunctional antibodies

Inclusion of adjacent loops can improve CDRH3 structural modeling

Inverse folding models trained on all proteins outperform those trained on Abs for CDR prediction

Kappa subtype restricts CDRH3 dynamics

Kinks in nanobody CDR3 loops are correctly predicted by structure prediction methods

Loops from one V-region chain affect the conformations and dynamics of loops in the other chain

Nanobody CDRH3 loops are longer but more compact

Optimal scaling of antibody LMs improves prediction of masked CDRH3 residues but not framework residues

QM-MM and unbiased MD are insufficient to correctly determine CDRH3 conformation

Template CDR retrieval using embeddings

pLDDT is inversely correlated with CDRH3 length

CDRH3 flanking residues more conserved and easier to predict than those in the center

Deep learning methods produce different CDRH3 conformers

Length independent clustering improves CDR prediction

Psi torsion angles are effective metadynamic CVs for CDRH3 and CDRL3

Sequence clusters and structural clusters of CDRH3 do not overlap

Complex prediction

Inverse folding models can predict antibody-antigen binding affinities with higher accuracy by correcting with predictions from unbound state

PAE weakly correlates with Ab-Ag binding

Structure-based methods outperform sequence-based methods on antigen-dependent antibody design

The structures of nanobodies with kinked CDR3 loops are more difficult to predict in complex with their antigen

Partial diffusion of de novo designed binders and VHHs improves refoldability and designability scores