Evolution-guided large language model is a predictor of virus mutation trends (preprint)

Emerging viral infections, especially the global pandemic COVID-19, have had catastrophic impacts on public health worldwide. The culprit of this pandemic, SARS-CoV-2, continues to evolve, giving rise to numerous sublineages with distinct characteristics. The traditional post-hoc wet-lab approach is lagging behind, and it cannot quickly predict the evolutionary trends of the virus while consuming high costs. Capturing the evolutionary drivers of virus and predicting potential high-risk mutations has become an urgent and critical problem to address. To tackle this challenge, we introduce ProtFound-V, an evolution-inspired deeplearning framework designed to explore the mutational trajectory of virus. Take SARS-CoV-2 as an example, ProtFound-V accurately identifies the evolutionary advantage of Omicron and proposes evolutionary trends consistent with wetlab experiments through in silico deep mutational scanning. This showcases the potential of deep learning predictions to replace traditional wet-lab experimental measurements. With the evolution-guided large language model, ProtFound-V presents a new state-of-the-art performance in key property predictions. Despite the challenge posed by epistasis to model generalization, ProtFound-V remains robust when extrapolating to lineages with different genetic backgrounds. Overall, this work paves the way for rapid responses to emerging viral infections, allowing for a plug-and-play approach to understanding and predicting virus evolution.

Running ahead of evolution - AI based simulation for predicting future high-risk SARS-CoV-2 variants (preprint)

The never-ending emergence of SARS-CoV-2 variations of concern (VOCs) has challenged the whole world for pandemic control. In order to develop effective drugs and vaccines, one needs to efficiently simulate SARS- CoV-2 spike receptor binding domain (RBD) mutations and identify high-risk variants. We pretrain a large pro- tein language model on approximately 408 million pro- tein sequences and construct a high-throughput screen- ing for the prediction of binding affinity and antibody escape. As the first work on SARS-CoV-2 RBD mu- tation simulation, we successfully identify mutations in the RBD regions of 5 VOCs and can screen millions of potential variants in seconds. Our workflow scales to 4096 NPUs with 96.5% scalability and 493.9X speedup in mixed precision computing, while achieving a peak performance of 366.8 PFLOPS (reaching 34.9% theo- retical peak) on Pengcheng Cloudbrain-II. Our method paves the way for simulating coronavirus evolution in or- der to prepare for a future pandemic that will inevitably take place.