A protein language model for exploring viral fitness landscapes
Jumpei Ito, Adam Strange, Wei Liu, Gustav Joas, Spyros Lytras, Keita Matsuno, Naganori Nao, Hirofumi Sawa, Keita Mizuma, Isshu Kojima, Jingshu Li, Tomoya Tsubo, Shinya Tanaka, Masumi Tsuda, Lei Wang, Yoshikata Oda, Zannatul Ferdous, Kenji Shishido, Takasuke Fukuhara, Tomokazu Tamura, Rigel Suzuki, Saori Suzuki, Shuhei Tsujino, Hayato Ito, Yu Kaku, Naoko Misawa, Arnon Plianchaisuk, Ziyi Guo, Alfredo A. Hinay, Kaoru Usui, Wilaiporn Saikruang, Keiya Uriu, Yusuke Kosugi, Shigeru Fujita, Jarel Elgin M.Tolentino, Luo Chen, Lin Pan, Wenye Li, Mai Suganami, Mika Chiba, Ryo Yoshimura, Kyoko Yasuda, Keiko Iida, Naomi Ohsumi, Shiho Tanaka, Kaho Okumura, Kazuhisa Yoshimura, Kenji Sadamas, Mami Nagashima, Hiroyuki Asakura, Isao Yoshida, So Nakagawa, Akifumi Takaori-Kondo, Kotaro Shirakawa, Kayoko Nagata, Ryosuke Nomura, Yoshihito Horisawa, Yusuke Tashiro, Yugo Kawai, Kazuo Takayama, Rina Hashimoto, Sayaka Deguchi, Yukio Watanabe, Yoshitaka Nakata, Hiroki Futatsusako, Ayaka Sakamoto, Naoko Yasuhara, Takao Hashiguchi, Tateki Suzuki, Kanako Kimura, Jiei Sasaki, Yukari Nakajima, Hisano Yajima, Takashi Irie, Ryoko Kawabata, Kaori Sasaki-Tabata, Terumasa Ikeda, Hesham Nasse, Ryo Shimizu, MST Monira Begum, Michael Jonathan, Yuka Mugita, Sharee Leong, Otowa Takahashi, Kimiko Ichihara, Takamasa Ueno, Chihiro Motozono, Mako Toyoda, Akatsuki Saito, Maya Shofa, Yuki Shibatani, Tomoko Nishiuchi, Jiri Zahradni, Prokopios Andrikopoulos, Miguel Padilla-Blanco, Aditi Konar, Kei Sato
Abstract
Successively emerging SARS-CoV-2 variants lead to repeated epidemic surges through escalated fitness (i.e., relative effective reproduction number between variants). Modeling the genotype–fitness relationship enables us to pinpoint the mutations boosting viral fitness and flag high-risk variants immediately after their detection. Here, we present CoVFit, a protein language model adapted from ESM-2, designed to predict variant fitness based solely on spike protein sequences. CoVFit was trained on genotype–fitness data derived from viral genome surveillance and functional mutation assays related to immune evasion. CoVFit successively ranked the fitness of unknown future variants harboring nearly 15 mutations with informative accuracy. CoVFit identified 959 fitness elevation events throughout SARS-CoV-2 evolution until late 2023. Furthermore, we show that CoVFit is applicable for predicting viral evolution through single amino acid mutations. Our study gives insight into the SARS-CoV-2 fitness landscape and provides a tool for efficiently identifying SARS-CoV-2 variants with higher epidemic risk. Ito et al. present CoVFit, an AI model that predicts variant fitness (transmissibility) from spike protein sequences alone. They further demonstrate its utility in forecasting viral evolution via single amino acid mutations.