Physics-driven structural docking and protein language models accelerate antibody screening and design for broad-spectrum antiviral therapy (preprint)

Therapeutic antibodies have become one of the most influential therapeutics in modern medicine to fight against infectious pathogens, cancer, and many other diseases. However, experimental screening for highly efficacious targeting antibodies is labor-intensive, which is exacerbated by evolving antigen targets under selective pressure such as fast-mutating viral variants. As a proof-of-concept, we developed a machine learning-assisted antibody generation pipeline that greatly accelerates the screening and re-design of immunoglobulins G (IgGs) against a broad spectrum of coronavirus variants. Using over 1300 IgG sequences derived from patient B cells bound with the viral spike's receptor binding domain (RBD), we first established protein structural docking models in assessing the IgG-RBD-ACE2 interaction interfaces and predicting their viral neutralizing activities with a confidence score. The confidence score is calculated as a fraction of IgG-blocking RBD binding sites versus all ACE2-interacting sites. Additionally, employing Gaussian process regression (also known as Kriging) in a latent space of an antibody language model, we predicted the IgGs' activity profiles against viral strains. Using functional analyses and experimental validations, we subsequently prioritized IgG candidates for neutralizing a broad spectrum of viral variants (wildtype, Delta, and Omicron) and preventing the infection of host cells in vitro and hACE2 transgenic mice in vivo. To further improve the blockade efficacy for the Delta strain (B.1.617), we rationally redesigned the IgG clones with single amino acid substitutions at the RBD-binding interface. Our work expedites applications of artificial intelligence in low-data regimes when limited data is available, including protein language models (using unlabeled data) and Kriging (using few labeled data) for antibody sequence analysis, activity prediction, and efficacy improvement, which are aided by physics-driven protein docking models for antibody-antigen interface structure analyses.

Efficient prevalence estimation and infected sample identification with group testing for SARS-CoV-2 (preprint)

The ongoing pandemic of SARS-CoV-2, a novel coronavirus, caused over 3 million reported cases of coronavirus disease 2019 (COVID-19) and 200,000 reported deaths between December 2019 and April 2020. Cases and deaths will increase as the virus continues its global march outward. In the absence of effective pharmaceutical interventions or a vaccine, wide-spread virological screening is required to inform where restrictive isolation measures should be targeted and when they can be lifted. However, limitations on testing capacity have restricted the ability of governments and institutions to identify individual clinical cases, appropriately measure community prevalence, and mitigate transmission. Group testing offers a way to increase efficiency, by combining samples and testing a small number of pools. Here, we evaluate the effectiveness of group testing designs for individual identification or prevalence estimation of SARS-CoV-2 infection when testing capacity is limited. To do this, we developed mathematical models for epidemic spread, incorporating empirically measured individual-level viral kinetics to simulate changing viral loads in a large population over the course of an epidemic. We used these to construct representative populations and assess pooling strategies for community screening, accounting for variability in viral load samples, dilution effects, changing prevalence and resource constraints. We confirmed our group testing framework through pooled tests on de-identified human nasopharyngeal specimens with viral loads representative of the larger population. We show that group testing designs can both accurately estimate overall prevalence using a small number of measurements and substantially increase the identification rate of infected individuals in resource-limited settings.