THE FUTURE OF SARS-COV-2 IN ECOLOGICAL COMMUNITIES

Aim 1. Predict future variants of SARS-CoV-2 and their zoonotic potential.While the scientific community quickly ramped up genomic surveillance since the pandemic began, most detected variants do not pose an ongoing threat to animals or to humans. Understanding which new variants have high zoonotic potential (the potential to infect both humans and animals) and what distinguishes them from low-risk variants, will be crucial to managing SARS-CoV-2. In Aim 1 we will extend our team's recently developed deep learning/machine learning algorithms to predict the sequence and structure of future betacoronaviruses with zoonotic potential, focusing on SARS-CoV-2 variants. We will leverage controllable (or controlled) generative AI (CGAI) and foundation modeling trained on large scale -omics data and their annotations. These represent leading edge AI research topics in which learning is supported by model training on a wider universe of foundational data underpinning complex patterns and biological relationships of interest (such as zoonotic potential arising from viral sequences and protein structures). CGAI developed through foundation modeling will generate a library of zoonotic variants and generative (evolutionary) trajectories that, when partnered with genomic surveillance, can be used as reference points to assess the zoonotic risk of SARS-CoV-2 variants, which continue to be detected and sequenced in the real world. Our project team has successfully advanced these methods in an adjacent domain to discover novel SARS-CoV-2 antiviral therapies and to predict zoonotic potential in viruses.Aim 2. Predict mammal hosts of SARS-CoV-2 variants.A common first step in assessing the risk posed by a novel variant is to experimentally characterize binding affinity between novel variants and mammal angiotensin converting enzyme-2 (ACE2) proteins, the receptor used by SARS-CoV-2 to enter host cells. This initial check is limited by the combinatorial enormity of potential variants and host species and the limited number of ACE2 sequences available for mammals (currently there are only 152 sequences available for over 6400 mammal species). We will leverage recent advances in protein representation learning models developed by our team to predict both the structure of mammalian ACE2 orthologs and their interactions with SARS-CoV-2 variants. We will extend two approaches in parallel: (1) a sequence-based approach, which uses a large-scale protein sequence model to predict binding affinity between novel viral variants (Aim 1) and mammal ACE2 structures (known and predicted); (2) a structure-based approach, which uses a structure-based predictor to estimate binding strengths of molecular conformations, namely the structure of the variants bound to known or predicted structures of ACE2 orthologs for all mammal species. Novel deep learning approaches offer computationally efficient and accurate methods for estimating protein-protein interactions, which will be passed on for experimental validations (cell entry assays, Aim 3).Aim 3. Validate the zoonotic potential for both variants and host species.Model predictions are made more accurate and more meaningful when partnered with empirical validation to iteratively improve the modeling process. We will link model outputs from Aims 1 and 2 with a leading edge pseudotype virus platform developed by members of our team. This platform measures the cell entry potential of any betacoronavirus, including any variant of SARS-CoV-2. These cost-effective, rapid, safe, and high-throughput assays will confirm cell entry and identify when variation in the SARS-CoV-2 receptor binding domains deviate from model predictions of zoonotic potential in humans and animals. We will test a subset of novel variants with high zoonotic potential from Aim 1 for cell entry in humans and a subset of mammal species, focusing on species whose ACE2 orthologs are predicted to support strong viral binding (Aim 2) and to be important SARS-CoV-2 hosts within the northeastern U.S. forest community (Aim 4).Aim 4. Develop community and spatial transmission theory for multi-host pathogens.?A theoretical framework for understanding the consequences of community ecology for zoonotic spread and spillover has been a longstanding challenge in disease ecology. Predicting the consequences of SARS-CoV-2 across multiple species requires going beyond predicting the susceptibility of individual host species, to understanding transmission dynamics in communities. We will examine how ecological interactions and spatial distributions determine SARS-CoV-2 prevalence in wildlife communities by developing theoretical and spatially explicit models of ecological interactions affecting SARS-CoV-2 dynamics. Transmission rates will be conditioned on space usage determined by the ecology of host species. This model will be instantiated for the northeastern U.S. forest community, which enables temporal matching among sequenced SARS-CoV-2 variants from ongoing wildlife surveillance in the U.S.