How to cluster structures based solely on 3Di

[luisbustom-oss]

Hi,

For a set of aminoacid sequences in a .fa file, I generated a db using Prost5.

foldseek createdb file.fa db --prostt5-model weights --gpu 1

I was also able to get the 3Di translations as .fasta file

foldseek lndb db_h db_ss_h
foldseek convert2fasta db_ss db_ss.fasta

However, when trying to cluster such structres...:
foldseek easy-cluster db cluster/cluster tmp --alignment-type 0

... it does not work:

Time for read in: 0h 0m 0s 210ms
tmp/2433386163849198049/clu_tmp/11610238211166517392/clustering.sh: line 124: 2435900 Segmentation fault      (core dumped) "$MMSEQS" clust "$INPUT" "${TMP_PATH}/pref_rescore1" "${TMP_PATH}/pre_clust" ${CLUSTER_PAR}
Error: Pre-clustering step died
Error: Search died

Could you please help me figure out how to proceed? I should say that I only want to cluster my structures on the basis of 3Di alignment, and not amino acid sequence.

Thank you in advance for your time

[milot-mirdita]

Would it be possible to upload an input set to reproduce the issue?

[luisbustom-oss]

Hi @milot-mirdita , thank you so much for your reply.

Here's the folder (with 1000 aminoacid sequences in "file.fa").

BTW I tried again:
foldseek easy-cluster db cluster/cluster tmp --alignment-type 0
and now got

createdb db tmp/7993952164266881713/input --gpu 0 --chain-name-mode 0 --db-extraction-mode 0 --distance-threshold 8 --write-mapping 0 --mask-bfactor-threshold 0 --coord-store-mode 2 --write-lookup 1 --input-format 0 --file-include '.*' --file-exclude '^$' --threads 40 -v 3 

Output file: tmp/7993952164266881713/input
[=================================================================] 100.00% 1.00K 0s 64ms   
Time for merging to input_ss: 0h 0m 0s 101ms
Time for merging to input_h: 0h 0m 0s 108ms
Time for merging to input_ca: 0h 0m 0s 105ms
Time for merging to input: 0h 0m 0s 393ms
No structures found in given input.
Error: query createdb died

1000.zip

[luisbustom-oss]

Hi @milot-mirdita , I was able to cluster the 3Di sequences using cluster instead of easy-cluster:
foldseek cluster db cluster tmp --alignment-type 0

I have a follow up question, though, as I would be interested in clustering the sequences based solely on 3Di-similarity, without taking into account the aminoacid sequence. I however noted that I get different results when clustering:

1. My original database, with the original aa sequences (file.fa) + their 3Di translations (3Di.tsv).
2. A masked database I created with the original 3Di translations (same 3Di.tsv) but with all aminoacid sequences being just VALs (file_masked.tsv). To do that:

• Fist create the masked db

foldseek tsv2db file_masked.tsv db_masked --output-dbtype 0
foldseek tsv2db 3Di.tsv db_masked_ss --output-dbtype 0
foldseek tsv2db header.tsv db_masked_h --output-dbtype 12

• Then cluster it like before:
  foldseek cluster db_masked cluster_masked tmp --alignment-type 0

As I mentioned pipelines 1 and 2 yield different results, so it seems to me like some aa information is still present. Could you help me understand at what stage of the procedure, and more generally if there is any way to cluster the structures based exclusively on 3Di and diregard the aminoacid composition entirely?

Thank you very much in advance for your time and this great tool,