EvoStruct: Bridging Evolutionary and Structural Priors for Antibody CDR Design via Protein Language Model Adaptation
arXiv:2605.21485v1 Announce Type: new Abstract: Equivariant graph neural network (GNN) methods for antibody complementarity-determining region (CDR) design achieve the highest sequence recovery but suffer from severe vocabulary collapse. The current best GNN methods over-predict very few amino acids, such as tyrosine and glycine, while ignoring functionally important residues. We trace this failure to GNN encoders learning amino acid distributions de novo from limited structural data, discarding substitution patterns encoded in evolutionary databases. To resolve this, we propose EvoStruct, whi
The increasing sophistication of protein language models (PLMs) and equivariant graph neural networks (GNNs) is enabling new approaches to complex biological design problems.
Improved antibody design techniques can significantly accelerate drug discovery and development for a wide range of diseases, leading to more effective and personalized therapies.
The ability to overcome limitations like vocabulary collapse in CDR design through hybrid evolutionary and structural approaches will lead to a broader range of viable antibody candidates.
- · Biopharmaceutical companies
- · Therapeutic development platforms
- · Patients with complex diseases
- · AI in drug discovery
- · Traditional antibody discovery methods
- · Companies relying on less efficient design processes
More efficient and effective antibody-based therapeutics can be developed and brought to market.
This could lead to a significant expansion in the pipelines of companies focused on immunology and oncology.
The success of EvoStruct might inspire similar hybrid approaches across other areas of protein engineering and synthetic biology, accelerating innovation further.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG