MSA filtering for protein-protein contact prediction

Question

[Omar Santiago]

Hello! I am attempting to extract a contact prediction between two proteins that have been shown to interact (but no structure has been determined). I am currently obtaining a list of sequences from ProteinA and ProteinB (separately) and generating a MSA for each individually. Should I filter my sequences or edit my MSA alignment in some way for DCA to work better between the two proteins? I ask because I don't know if I should use trimAl or try to remove sequences 90% redundant.

Answer 1

[Mensur Dlakic]

It would be helpful to tell us what is the prediction method you plan to use. But generally speaking, you don't need to do any trimming or redundancy removal, as most DCA methods know how to weigh sequences. The most important thing for good contact prediction is to get a deep and diverse alignment. There is no hard definition of what a deep alignment is, but for practical purposes let's say >1000 sequences. Depending on proteins of interest, it may be a good idea to search through metagenomic and environmental databases.

[Omar Santiago]

Thank you! For the prediction method, I am planning on using plmDCA (https://github.com/pagnani/PlmDCA). I'll try to find many sequences for the alignment then!