I have the coding sequences of a WT gene and several mutants of this gene (deletion mutations varying from 5bp to 50% of the sequence deleted). What is the best method for inferring the impact of these mutations on the final protein?

I am looking at AlphaFold but papers have found it is no good at predicting protein stability change (ΔΔG) based on point mutations, and AlphaFold answer one of their FAQs by saying "AlphaFold has not been validated for predicting the effect of mutations".

So... what is the best method for predicting effects of mutations based on sequences?

  • $\begingroup$ Not sure, but look into Rosetta or DynaMut. Obviously, actual experimental validation is the best. Are there known deletions in related species or the same species (ExAC, gnomAD, ClinVar)? $\endgroup$
    – BigMistake
    Feb 22 at 20:02

1 Answer 1


I would suggest looking at the sequence conservation of the protein domains / sequence positions of interest.

One such tool that is oriented towards human genomics is CADD. This tool can provide a score predicting how damaging a mutation at the position in question might be.

But in the absence of a convenient interface, you could generate a sequence alignment yourself of homologous proteins and compute the conservation (similarity or identity) of the position across such homologous proteins.

Unfortunately I am not sure that these methods will make you feel much wiser, they will largely be useful for saying that one area of a protein is more important than another, they will not necessarily tell you much about the functional changes in the protein.

It is of course always a good idea to look at the domain annotations of the protein in question, e.g. mutations in the active site of an enzyme may affect activity, mutations in a DNA binding domain are more likely to affect DNA affinity, etc.

Uniprot is one place where you can investigate both of these at once to some extent.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .