Summary
Protein structure prediction methods reliant on MSAs outperform those that use PLMs. This was observed in CASP15 and documented by (1,2).
Details
This may stem from the superior performance of these methods on contact prediction (documented by the authors of MSA Transformer; (3)) and structure interface scoring (4).
1.
Hu M, Yuan F, Yang K, Ju F, Su J, Wang H, et al. Exploring evolution-aware & -free protein language models as protein function predictors. Advances in Neural Information Processing Systems. 2022;35:38873–84. Available from: https://papers.nips.cc/paper_files/paper/2022/hash/fe066022bab2a6c6a3c57032a1623c70-Abstract-Conference.html
2.
Barrett TD, Villegas-Morcillo A, Robinson L, Gaujac B, Adméte D, Saquand E, et al. So ManyFolds, So Little Time: Efficient Protein Structure Prediction With pLMs and MSAs. openRxiv; 2022. Available from: https://doi.org/10.1101/2022.10.15.511553
3.
Rao R, Liu J, Verkuil R, Meier J, Canny JF, Abbeel P, et al. MSA Transformer. openRxiv; 2021. Available from: https://doi.org/10.1101/2021.02.12.430858
4.
Liu D, Zhang B, Liu J, Li H, Song L, Zhang G. Assessing protein model quality based on deep graph coupled networks using protein language model. Briefings in Bioinformatics. 2023;25(1). Available from: https://doi.org/10.1093/bib/bbad420