Method’s application to antibody Rep-seq data.

(a) Depicts the inference of the selection process, where the initial antibody population represents the unimmunized repertoire (negative set). After the immune response, the library undergoes selection to bind the antigen (positive set), shaping the immunoglobulin population. (b) Displays a plot of background and selection energies for both negative and positive sets. Red crosses represent the test set of antibodies with affinity measures (panel e). The selection energy effectively distinguishes antibodies in the immunized repertoire from those in the unimmunized repertoire. (c) Classification Task: Demonstrates the model’s ability to discriminate between binders and non-binders by presenting results on a random test set composed of negative and positive antibodies. ROC curves for the GPI and TT cases yield area under the curve (AUC) values of 0.89 and 0.98, respectively. (d) and (e) Model energy vs. Experimental Affinity: Show scatter plots comparing the selection energy (y-axis in panel b) with affinity measures for a set of antibodies. Specifically, EC50 values for TT and K_d measures for GPI. Notably, a significant correlation (Spearman coefficient 0.76) is observed in the latter case. The GPI test set is indicated by red crosses in panel (b).