32nd International Dynamics & Evolution of Human Viruses

Molecular design of more broadly neutralizing antibodies against HIV-1 using reliable machine learning models

Not scheduled

20m

Abbaye de Royaumont, Asnières-sur-Oise, France

Poster Vaccines & immune escape Virtual posters

Speaker

Aime Bienfait Igiraneza (University of Oxford)

Description

The discovery of very potent, broadly neutralizing antibodies (bnAbs) against HIV-1 has revived the hope for new treatment and prevention methods against the virus. Nevertheless, several clinical trials have shown that the breadth of such bnAbs is still limited by viral resistance. Here, we explore whether currently available neutralization
datasets on bnAbs predicted to bind similar epitopes are sufficient to predict mutations in existing HIV-1 bnAbs that will increase their breadth without compromising potency. We propose a framework for reliably training and evaluating machine learning models that predict HIV-1 resistance to bnAb mutants without mutant-specific neutralization
data. Using such models along with VRC01 and VRC07 as reference bnAbs, we identify mutants with a structural basis for increased breadth. In addition to the discovery of novel potent bnAbs against HIV-1, our supervised machine learning protocol should extend to increase the breadth of bnAbs against highly diverse pathogens.