# Residue coevoution from position wise residue preferences rather than multiple sequence alignments

Hello experts in residue coevolution analysis of proteins,

I am looking for a way to calculate residue coevolution (coupling scores) between all pairs of residues of a protein. The issue is that rather than having a Multiple Sequence Alignment (MSA) which is used as an input by the relevant tools (e.g. evcouplings), I have position wise preferences for 20 amino acids for the protein.

For the protein with length L, I have a 20*L matrix of position-wise preferences for amino acids. I would like to use this matrix as a direct input for coevolution analysis.

Would it be possible to calculate the residue coevolution direcly from a residue preference matrix?

Any recommendation for a method of choice or a software tool would be helpful.

• Coevolution requires you to know when you have residue X at position n you will preferentially have Y at m. Namely the covariance of n with m. If you have a PSSM matrix like you have, you cannot infer the covariance. You would need a L*L*20*20 tensor. Jun 5 '20 at 12:13
• However, your PSSM-like matrix contains useful information, such as what is core and what is surface. But I am not sure what tool uses it. Have a gander at predictioncenter.org/casp14/index.cgi Jun 5 '20 at 12:16