Question: Does foldseek work well on structures with no aa sequences?

[ML-J]

Greetings team! We are using foldseek for clustering on a batch of pdbs without aa sequence. We set the alignment-mode to 3di-alignment. But the result still shows that the pdbs in one cluster have low similarities( rmsd ~9) between each other.

The command we used is foldseek easy-cluster ./data cluster0.8 tmp -c 0.8 --alignment-mode 0

We tried series of optional parameters combinations with little success. Is foldseek suitable for clustering non-sequence strcutures? And how should we adjust the parameters to obtain better cluster results?

[martin-steinegger]

What kind of data do you try to cluster? You could try to set a --tmscore-threshold. But just a disclaimer, foldseek is not not meant to cluster huge set of near identical structures but rather sets of multiple groups of protein structures.

[ML-J]

Thanks for your reply! Concretely, our structure files are obtained from generative models. The pdb files only contain the backbone atom coordinates and the residue types are all ALA.