Proteins are fundamental to life. The full extent of their biology is beyond our ability to characterize with experimental approaches in the physical laboratory. Accurate digital representations could accelerate the discovery of protein biology through virtual experiments. We propose language modeling to learn unified and general representations that can be scaled to all of protein biology. Building on these representations, we develop a structure prediction model that exceeds the performance of established methods for biomolecular complex prediction across benchmarks, including for the interactions of antibodies with their targets. A simple search procedure yields high experimental success rates for the discovery of proteins with nanomolar binding affinities for both miniproteins and single-chain antibodies, a modality critical for therapeutic design. Study of the concepts in the language model's representation space reveals a systematic organization aligned with the reductionist understanding of proteins developed through empirical science. Leveraging this organization, we generate a comprehensive map of protein biology encompassing over 6.8 billion sequences and 1.1 billion predicted structures, identifying connections across known and unknown biology. As a whole, this shows language modeling as a powerful substrate for representing the biology of proteins, operating across scales from the prediction and design of protein interactions at the atomic level, to identifying properties of proteins at different levels of granularity and abstraction, to the scale of mapping connections between proteins across billions of years of evolution.
During the COVID-19 pandemic, SARS-CoV-2 infected millions of people and claimed hundreds of thousands of lives. Virus entry into cells depends on the receptor binding domain (RBD) of the SARS-CoV-2 spike protein (S). Although there is no vaccine, it is likely that antibodies will be essential for protection. However, little is known about the human antibody response to SARS-CoV-2 1–5 . Here we report on 149 COVID-19 convalescent individuals. Plasmas collected an average of 39 days after the onset of symptoms had variable half-maximal neutralizing titers ranging from undetectable in 33% to below 1:1000 in 79%, while only 1% showed titers >1:5000. Antibody cloning revealed expanded clones of RBD-specific memory B cells expressing closely related antibodies in different individuals. Despite low plasma titers, antibodies to three distinct epitopes on RBD neutralized at half-maximal inhibitory concentrations (IC 50 s) as low as single digit ng/mL. Thus, most convalescent plasmas obtained from individuals who recover from COVID-19 do not contain high levels of neutralizing activity. Nevertheless, rare but recurring RBD-specific antibodies with potent antiviral activity were found in all individuals tested, suggesting that a vaccine designed to elicit such antibodies could be broadly effective.
Science frequently benefits from teams of interdisciplinary researchers. However, most scientists don’t have access to experts from multiple fields. Fortunately, large language models (LLMs) have recently shown an impressive ability to aid researchers across diverse domains by answering scientific questions. Here, we expand the capabilities of LLMs for science by introducing the Virtual Lab, an AI-human research collaboration to perform sophisticated, interdisciplinary science research. The Virtual Lab consists of an LLM principal investigator agent guiding a team of LLM agents with different scientific backgrounds (e.g., a chemist agent, a computer scientist agent, a critic agent), with a human researcher providing high-level feedback. We design the Virtual Lab to conduct scientific research through a series of team meetings, where all the agents discuss a scientific agenda, and individual meetings, where an agent accomplishes a specific task. We demonstrate the power of the Virtual Lab by applying it to design nanobody binders to recent variants of SARS-CoV-2, which is a challenging, open-ended research problem that requires reasoning across diverse fields from biology to computer science. The Virtual Lab creates a novel computational nanobody design pipeline that incorporates ESM, AlphaFold-Multimer, and Rosetta and designs 92 new nanobodies. Experimental validation of those designs reveals a range of functional nanobodies with promising binding profiles across SARS-CoV-2 variants. In particular, two new nanobodies exhibit improved binding to the recent JN.1 or KP.3 variants of SARS-CoV-2 while maintaining strong binding to the ancestral viral spike protein, suggesting exciting candidates for further investigation. This demonstrates the ability of the Virtual Lab to rapidly make impactful, real-world scientific discovery.
Serology has provided valuable diagnostic and epidemiological data on antibody responses to SARS-CoV-2 in diverse patient cohorts. Deployment of high content, multiplex serology platforms across the world, including in low and medium income countries, can accelerate longitudinal epidemiological surveys. Here we report multiSero, an open platform to enable multiplex serology with up to 48 antigens in a 96-well format. The platform consists of three components: ELISA-array of printed proteins, a commercial or home-built plate reader, and modular python software for automated analysis (pysero). We validate the platform by comparing antibody titers against the SARS-CoV-2 Spike, receptor binding domain (RBD), and nucleocapsid (N) in 114 sera from COVID-19 positive individuals and 87 pre-pandemic COVID-19 negative sera. We report data with both a commercial plate reader and an inexpensive, open plate reader (nautilus). Receiver operating characteristic (ROC) analysis of classification with single antigens shows that Spike and RBD classify positive and negative sera with the highest sensitivity at a given specificity. The platform distinguished positive sera from negative sera when the reactivity of the sera was equivalent to the binding of 1 ng mL −1 RBD-specific monoclonal antibody. We developed normalization and classification methods to pool antibody responses from multiple antigens and multiple experiments. Our results demonstrate a performant and accessible pipeline for multiplexed ELISA ready for multiple applications, including serosurveillance, identification of viral proteins that elicit antibody responses, differential diagnosis of circulating pathogens, and immune responses to vaccines.
The sequencing of antibody repertoires of B-cells at increasing coverage and depth has led to the identification of vast numbers of immunoglobulin heavy and light chains. However, the size and complexity of these Adaptive Immune Receptor Repertoire sequencing (AIRR-seq) datasets makes it difficult to perform exploratory analyses. To aid in data exploration, we have developed AIRRscape, an R Shiny-based interactive web browser application that enables B-cell receptor (BCR) and antibody feature discovery through comparisons among multiple repertoires. Using AIRR-seq data as input, AIRRscape starts by aggregating and sorting repertoires into interactive and explorable bins of germline V-gene, germline J-gene, and CDR3 length, providing a high-level view of the entire repertoire. Interesting subsets of repertoires can be quickly identified and selected, and then network topologies of CDR3 motifs can be generated for further exploration. Here we demonstrate AIRRscape using patient BCR repertoires and sequences of published monoclonal antibodies to investigate patterns of humoral immunity to three viral pathogens: SARS-CoV-2, HIV-1, and DENV (dengue virus). AIRRscape reveals convergent antibody sequences among datasets for all three pathogens, although HIV-1 antibody datasets display limited convergence and idiosyncratic responses. We have made AIRRscape available as a web-based Shiny application, along with code on GitHub to encourage its open development and use by immuno-informaticians, virologists, immunologists, vaccine developers, and other scientists that are interested in exploring and comparing multiple immune receptor repertoires.