Protein models for "Towards a structurally resolved human protein interaction network"

dataset

posted on 2021-11-23, 05:10 authored by Arne ElofssonArne Elofsson, Petras Kundrotas, Pedro Beltrao, Patrick BryantPatrick Bryant, Gabriele PozzatiGabriele Pozzati

Dataset description

This is all prediction associated with the paper entitled 'Towards a structurally resolved human protein interaction network'.

Predicted protein models for 65000 human protein-protein interactions. Five datasets pairs: HuRUm HuMap2, Random single chain: HuRI and HuMap.

Additional resources can be found here: https://archive.bioinfo.se/huintaf2/

Protein interaction data and annotations

Human protein pairs known to physically interact were obtained from the Hu.MAP dataset, retaining pairwise interactions with >=0.5 confidence, and most interactions from the HuRI dataset.

Protein complex prediction

To predict protein complexes of pairwise interactions, we used the FoldDock pipeline (Bryant et al., 2021) based on AlphaFold2 (Jumper et al., 2021). We use the option of fused+paired multiple sequence alignments (MSAs) and run the model configuration m1-10-1 as this provides the highest success rate accompanied by a 20-fold speed-up. Both the fused and paired MSAs are constructed from running HHblits on every single chain against Uniclust30. The fused MSA is generated by simply concatenating the output of each of the single-chain HHblits runs for two interacting chains. The paired MSA is constructed by combining the top hit for each matching OX identifier between two interacting chains, using the output from the single-chain HHblits runs.