Abstract
Generation of de novo peptides docking snake three-fold α-neurotoxins (3Ftx) were designed using computational diffusion models as alternatives to screening or designing larger antibody/protein-based and smaller drug-like anti-venoms. To approach the high variability of snake venoms, the previously described α-neurotoxin consensus amino acid sequence (ScNtx) was selected as the model to explore. Among the thousands of RFdiffusion models randomly generated, a few peptide candidates predicting low picoMolar affinities, ~ 30-mer sizes and targeting all ScNtx finger-loops (I, II and III) were discovered. These new individual or mixed candidates, constitute examples for experimental evaluations, because such small peptides could be easily produced by recombinant bacteria and/or chemical synthesis. Similar computational strategies could also be extended to other snake venom consensus.
Supplementary materials
Title
GraphycalAbstract.pse cartoons of snake-venom and anti-peptide designed
Description
Grey cartoon, carbon backbone of ScNtx consensus snake venom (7luw.pdb). Red cartoon, carbon backbone of peptide 11.4n1 designed by RFdiffusion simultaneously targeting dipeptides, gap-filled with GG (yellow cartoons) and including the A1G mutation. This file also includes the 7luw complexes predicted by Alphafold2 of 11.4n1 including several individual mutations and Prodigy predicted affinities:
9bk7, de novo anti-ScNtx protein10 (cyan) Prodigy predicting 34 mM affinity,
11.4n1 noGs (yellow) Prodigy predicting 258 pM affinity,
11.4n1+GGGGG (salmon), Prodigy predicting 40 pM affinity
11.4n1+GG (chocolate), Prodigy predicting 39 pM affinity,
11.4n1+GG+A1G (red), Prodigy predicting 0.04 pM affinity,
11.4n1+GG+A1G+A2G (orange), Prodigy predicting 257 pM affinity,
11.4n1+GG+A12L (violet), Prodigy predicting 1 pM affinity,
11.4n1+GG+A1G+A2G+A12L (green), Prodigy predicting 2444 pM affinity,
Actions
Supplementary weblinks
Title
RFdiffusion. From images to proteins
Description
A short divulgative video briefly explaining the results of applying RFdiffusion to de novo design of proteins, based on similar noising/de-noising images relying in training deep-learning networks on 3D pdb protein structures rather than images. The video shows random generated peptides trying to fit the target protein surface.
Actions
View