Summary
The Invariant point attention module from Alphafold2 is not mandatory for protein structure prediction (1). In (2), ablation studies show only a small drop in performance, while in RosettaFold, its replacement with an SE3-transformer led to no drop in accuracy. (2) found that removing IPA led to very minor decreases in structure prediction quality, unless recycling was also removed. However, (3) found when IPA was replaced with a single linear layer, the Evoformer lost 80% of its ability to predict secondary structure.
Figures
Ref (2)
1.
Baek M, Anishchenko I, Humphreys IR, Cong Q, Baker D, DiMaio F. Efficient and accurate prediction of protein structure using RoseTTAFold2. openRxiv; 2023. Available from: https://doi.org/10.1101/2023.05.24.542179
2.
Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, et al. Highly accurate protein structure prediction with AlphaFold. Nature. 2021;596(7873):583–9. Available from: https://doi.org/10.1038/s41586-021-03819-2
3.
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