Protein language models, or PLMs, have been quite successful in understanding proteins. However, they struggle with antibodies, especially their hypervariable regions, which don't follow the usual evolutionary rules. Researchers have come up with a new method called Antibody Mutagenesis-Augmented Processing, or AbMAP, to tackle this issue. AbMAP tweaks these general models to better handle antibody sequences by focusing on their structure and how they bind to specific targets. This approach lets the models accurately predict how changes in the antibody sequence affect its ability to bind to antigens and identify its specific binding sites.

To test AbMAP, scientists used it to improve a set of antibodies that target a SARS-CoV-2 peptide. The results were impressive: an 82% success rate and an increase in binding strength by up to 22 times. Not only that, but AbMAP also helps in analyzing large groups of immune receptors. It turns out that while the sequences of B-cell receptors vary greatly between individuals, they tend to have similar structural and functional features. This method can keep up with future advancements in foundational PLMs. The hope is that AbMAP will speed up the design and modeling of antibodies, help find new antibody-based treatments faster, and deepen our understanding of the immune system.

Boosting Antibody Modeling with Transfer Learning

Actions