Summary
Running Inverse folding in a random order outperforms fixed order (1). A related scheme is order-agnostic autoregressive diffusion, used by a sequence diffusion model, in which the identities of individual residues are determined one at a time. The forward noising process masks one residue at a time, in effect. This was shown by (2) to outperform whole-sequence simultaneous diffusion. Applying this scheme on diffusion models for Multiple sequence alignments outperforms MSA Transformer.
1.
Dauparas J, Anishchenko I, Bennett N, Bai H, Ragotte RJ, Milles LF, et al. Robust deep learning–based protein sequence design using ProteinMPNN. Science. 2022;378(6615):49–56. Available from: https://doi.org/10.1126/science.add2187
2.
Alamdari S, Thakkar N, van den Berg R, Tenenholtz N, Strome R, Moses AM, et al. Protein generation with evolutionary diffusion: sequence is all you need. openRxiv; 2023. Available from: https://doi.org/10.1101/2023.09.11.556673