AF3/Boltz2 and HHBlitz/mmseqs2

[AlexanderMath]

Did anyone compare LIS between AF3/Boltz2 and hhblitz/mmseqs2?

I tried one PPI where af3, boltz2+af3msa, boltz2+mmseqs2 all give the same structure, but LIS_af3<0.2 and LIS_boltz2_mmseqs>0.2.

[flyark]

I found that AlphaFold-Multimer often performed better than AF3 in my tested datasets (challenging candidates with a lot of flexible region).

Depending on the MSA quality and depth in the same prediction tools, the confidence for predicted structures can differ. Within the same tools and settings, it would be nice to compare with control proteins such as GFP.

This paper (https://academic.oup.com/bib/article/26/2/bbaf186/8120239) tested GPCR structure benchmarks and compared several prediction tools including AFM, AF3, and Chai-1. But I am not sure whether they tested different MSA sets.