Fig. 1. RAIN pipeline for automatic identification of bNAbs.

Data collected from the CATNAP database (bNAbs) and healthy donor repertoires (mAbs) were converted into a features table to train and validate four machine-learning models: anomaly detection (AD), Decision Tree (DT), random forest (RF), and super learner (SL). We performed single-cell BCR sequencing from HIV-1 seropositive donors with sera exhibiting broadly neutralizing activities (illustrated by the brown arms) or without neutralizing activities (illustrated by the white arms). BCR sequences were processed by the Immcantation workflow and analyzed as a feature table. Next, the predicted bNAbs found by the four algorithms were produced and tested in neutralization and binding assays. The image was created using BioRender.com.