Fig 1.
Prediction performance of the classifiers.
The AUC performances for each bNAb classifier using the oligo kernel were determined by 10 runs of a stratified 5-fold nested cross-validation. On the x-axis, the different classifiers are presented, named according to the bNAb they are trained on. The colors of the boxes refer to the epitope category of the corresponding bNAb. The prediction performance of a random classifier is depicted by the gray dashed line.
Table 1.
Final parameter settings for the oligo kernel classifiers for each bNAb.
Table 2.
Learnt discriminant signals by each bNAb classifier that are supported by literature.
Fig 2.
Motif logo for the test sequence HM469973 using the PG9 classifier.
For the motif logo the contribution of 5% of the strongest discriminant signals to the classification is considered. The height of the letters depends on the proportional contribution to the classification. Amino acids of the test sequence that influence the classification outcome towards neutralization susceptibility are displayed in capital letters and blue color; lowercase letters and orange color if they contribute to neutralization resistance. For better interpretability, the corresponding positions of the amino acids in the envelope sequence of the HIV strain HXB2 are shown on the x-axis.
Fig 3.
Characteristics of the non-B subtype HIV variants over time.
Predicted neutralization sensitivity of non-B subtype HIV-1 variants to bNAbs over time. A - D: the predicted neutralization sensitivity of HIV-1 samples of the non-B subtype to the bNAbs 3BNC117, 35O22, PGT121, and PG16. E: the ratio of R5-tropic and X4-capable non-B subtype HIV-1 variants in the Los Alamos HIV sequence database over the six time periods. F: predicted neutralization sensitivity of R5-tropic variants of the non-B subtype to PG16.
Fig 4.
Association between coreceptor usage and neutralization sensitivity.
Relative number of resistant (orange) and susceptible (blue) strains with regard to their coreceptor usage for the bNAbs PGT128 and VRC-PG04, as well as for VRC01, PG9, and PG16. Statistical significance was assessed with a two-sided Fisher’s exact test. Data for VRC01, VRC-PG04, PG9 and PG16 was taken from Pfeifer et al. [63].