Missense variant proteins 3D Structure prediction

Question

[Esraa]

Greetings!

I normally use Alphafold2 for protein 3D structure modeling, but i recently found out that it is insensitive for predicting structures that only differ from the wild type by a missense variant.

So my question is, what 3D structure modeling tool can i use for predicting the structures of variant proteins (MIssense variants)?

Answer 1

[Mensur Dlakic]

I normally use Alphafold2 for protein 3D structure modeling, but i recently found out that it is insensentive for predicting sturctures that only differ from the wild type by a missense variant.

That's because in most cases a single mutation will have a small effect on protein structure, at least the way it looks globally. A mutant may "look" the same but be different energetically.

If you are starting from a known wild-type structure, Rosetta and FoldX have dedicated protocols that will do the modeling much faster than AlphaFold2. The main difference one should expect here is in folding energies, specifically the delta-deltaG of mutants compared to the wild-type.

• https://github.com/JinyuanSun/DDGScan
• https://foldxsuite.crg.eu/

[Esraa]

This actually helps a lot, the folding energies aspect is truly important and i was going to search for tools to calculate it after i finish modeling. so those protocols sum it up.

I will be looking into them and update here if i reach something new. Thank you so much!

Answer 2

[dthorbur]

If you are working within academia you can use AlphaFold3, though in it's current form you will be limited to only 20 jobs per day and can only use the web server. Whilst I haven't seen a benchmark, it often performs significantly better than AF2.

If you are limited to AF2, you can use AlphaFold-Missense (paper here).

Another possibility, and again behind a non-commercial use license is ESM3. This only came out recently, and I haven't seen many results yet.

I am unfamiliar with tools outside of AF-derivates for effective single variant structural modelling, but I'm hoping ESM3 works well enough.

[Esraa]

Yes i have seen Alpha-Missense but it is more of a variant effect prediction database than a 3D Structure modeling tool, it is really helpful for providing pathogenecity information though. Thank you so much for helping!