Machine Learning Approach Identifies HIV-1 Broadly Neutralizing Antibodies
A computational method has been developed to identify HIV-1 broadly neutralizing antibodies (bNAbs) from non-selected immune repertoires, researchers reported in the journal Nature Communications.
“To date, the identification of bNAbs has required B-cell isolation and clonal expansion from selected individuals possessing sera with broadly neutralizing activity. This step is followed by antibody cloning and experimental validation of their neutralization potential,” explained corresponding author Laurent Perez, of the Lausanne University Hospital and University of Lausanne in Switzerland, and coauthors in the study introduction. “While both steps represent an important research effort, the process has benefited from identified immune donors.”
The Rapid Automatic Identification of bNAbs from Immune Repertoires (RAIN) approach is a machine-learning method. Unlike approaches that use one-hot encoding amino acid sequences or structural alignment for prediction, RAIN combines several selected sequence-based features, the team reported.
In a validation analysis that used previously identified bNAbs, RAIN had a 100% prediction accuracy and an area under the curve value that ranged from 0.92 to 1, according to the paper.
To demonstrate RAIN’s performance, researchers isolated class-switched memory B cells from HIV-1 immune donors and performed single-cell B-cell receptor sequencing. The process allowed them to identify three bNAbs, none of which was found in immune donors without broadly neutralizing activities.
“In summary, our approach offers an innovative, straightforward method to search and identify antibodies in immune repertoires, accelerate antibody discovery, and might shed light on potentially unexplored mechanisms of HIV-1 immune escape,” researchers wrote.
Reference
Foglierini M, Nortier P, Schelling R, et al. RAIN: machine learning-based identification for HIV-1 bNAbs. Nat Commun. 2024;15(1):5339. doi:10.1038/s41467-024-49676-1