By allowing protein language models (PLMs) to learn from each other’s most confident predictions, we compressed the collective knowledge of existing PLMs into VESM — a single sequence-only model that outperforms state-of-the-art hybrid methods. VESM predictions extended beyond binary pathogenicity classification, accurately quantifying the severity of variant effects on clinical phenotypes.
This is a preview of subscription content, access via your institution
Access options
Access Nature and 54 other Nature Portfolio journals
Get Nature+, our best-value online-access subscription
$32.99 / 30 days
cancel any time
Subscribe to this journal
Receive 12 print issues and online access
$259.00 per year
only $21.58 per issue
Buy this article
- Purchase on SpringerLink
- Instant access to the full article PDF.
USD 39.95
Prices may be subject to local taxes which are calculated during checkout
References
Rives, A. et al. Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences. Proc. Natl Acad. Sci. USA 118, e2016239118 (2021). This paper introduced the ESM family and showed that protein language models can learn rich biological constraints from sequence data.
Brandes, N. et al. Genome-wide prediction of disease variant effects with a deep protein language model. Nat. Genet. 55, 1512–1522 (2023). This paper introduced protein language models as foundational tools for genome-wide variant effect prediction.
Cheng, J. et al. Accurate proteome-wide missense variant effect prediction with AlphaMissense. Science 381, eadg7492 (2023). This paper introduced AlphaMissense as a state-of-the-art hybrid variant effect predictor trained on sequence alignments, 3D protein structures and population allele frequency data.
Karczewski, K. J. et al. Systematic single-variant and gene-based association testing of thousands of phenotypes in 394,841 UK Biobank exomes. Cell Genomics 2, 100168 (2022). This paper introduces Genebass, a public resource of rare variant association statistics from UK Biobank exomes, providing the data we used to evaluate VESM predictions.
Notin, P. et al. ProteinGym: large-scale benchmarks for protein fitness prediction and design. In Advances in Neural Information Processing Systems Vol. 36 (eds. Oh, A. et al.) 64331–64379 (Curran Associates, 2023). This article introduces ProteinGym, a comprehensive benchmark of clinical and deep-mutational-scanning datasets that enables systematic comparison of variant effect prediction methods.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This is a summary of: Dinh, T., Jang, S. K., Zaitlen, N. & Ntranos, V. Compressing the collective knowledge of ESM into a single protein language model. Nat. Methods https://doi.org/10.1038/s41592-026-03050-9 (2026).
Rights and permissions
About this article
Cite this article
Bridging the gap between hybrid and sequence-only protein language models. Nat Methods (2026). https://doi.org/10.1038/s41592-026-03049-2
Published:
Version of record:
DOI: https://doi.org/10.1038/s41592-026-03049-2
