Foldseek TMAlign option alntmscore inconsistent with tmalign server results.

[WesleyMBotelloSmith]

Expected Behavior

foldseek run with tmalign mode (--alignment-option 1) should return the same (or similar) tmscore to running tmalign by hand or via tmalign server

Current Behavior

tmscore returned is sometimes drastically different

Steps to Reproduce (for bugs)

see colab notebook -> downloads and runs foldseek, writes needed pdb files (prints atom record text to files): https://colab.research.google.com/drive/1fXGBusa5HoGspaT_HgKS7i90XOa5vvKR?usp=sharing

tmalign server (https://zhanggroup.org/TM-align/), copy atom records from 'synthetic_pdb.pdb' and paste into box for input_1 and copy '1vh6_B.pdb' atom records and paste as input_2 (see colab notebook text cells for atom record text ).

Foldssek Output (for bugs)

Please make sure to also post the complete output of Spacepharer. You can use gist.github.com for large output. Alignment file: synthetic_pdb.pdb 1vh6_B.pdb 9.164E-01

foldseek output gist url: https://gist.github.com/WesleyMBotelloSmith/5df62e3a7cb91ad1dcdcc50e57c9ba25

Context

Foldseek is being used to provide a similarity metric (tmalign mode tmscore) for assessing the diversity and novelty of a set of pdbs created by a generated model.

Your Environment

Include as many relevant details about the environment you experienced the bug in. -standard google colab notebook foldseek install command: !wget https://mmseqs.com/foldseek/foldseek-linux-avx2.tar.gz; tar xvzf foldseek-linux-avx2.tar.gz; export PATH=$(pwd)/foldseek/bin/:$PATH

tmalign server used to compute reference tmalign based tmscores

• Git commit used (The string after "MMseqs Version:" when you execute foldseek without any parameters):
• Which foldseek version was used (Statically-compiled, self-compiled, Conda, etc.):
• For self-compiled and Homebrew: Compiler and Cmake versions used and their invocation:
• Server specifications (especially CPU support for AVX2/SSE and amount of system memory):
• Operating system and version:

[ekiefl]

I've also observed this. Can anyone shed some light on this?

[dpretorius]

Is there any update on this? My local Foldseek with tmalign mode (--alignment-option 1) is returning different hits than the server. The top hit returned locally is not even present for all hits on the server.