Undefined bit score (-2147483648)?

[Andy-B-123]

Hi, I'm running Foldseek with some structures predicted using esmfold and for some hits I'm getting a negative/undefined bit score?

I am using the SwissProt database generated using the 'databases' command from foldseek. I'm using the latest version and this only happens for 1 of the ~30,000 structures I've generated, but it seems to happen for a range of hits with e-values and I normally filter on bits so this would be removing potentially real hits! I'm not sure if it's related to the protein sequence/structure or the search. I have attached the .pdb file here.

Thank you for having a look!

$ grep '\-2147483648' pgaptmp_006197.foldseek.out
#query	target	fident	alnlen	mismatch	gapopen	qstart	qend	tstart	tend	evalue	bits
pgaptmp_006197	AF-Q45827-F1-model_v4	0.128	749	513	0	259	1007	9	597	2.579E-13	-2147483648
pgaptmp_006197	AF-P07168-F1-model_v4	0.142	985	693	0	199	1183	21	829	7.470E-20	-2147483648
pgaptmp_006197	AF-P07167-F1-model_v4	0.137	987	703	0	198	1184	20	835	2.069E-17	-2147483648
pgaptmp_006197	AF-A7MRY4-F1-model_v4	0.105	1109	739	0	224	1332	9	835	4.819E-15	-2147483648
pgaptmp_006197	AF-Q9KLK7-F1-model_v4	0.147	1208	719	0	124	1331	2	845	1.331E-25	-2147483648
pgaptmp_006197	AF-P58356-F1-model_v4	0.177	1241	744	0	90	1330	4	908	4.257E-20	-2147483648
pgaptmp_006197	AF-P39453-F1-model_v4	0.185	1145	680	0	91	1235	5	839	9.779E-21	-2147483648
pgaptmp_006197	AF-Q55445-F1-model_v4	0.138	726	534	0	241	861	198	923	3.166E-03	-2147483648
pgaptmp_006197	AF-Q5A599-F1-model_v4	0.337	1240	627	0	93	1332	53	999	9.278E-51	-2147483648
pgaptmp_006197	AF-Q86CZ2-F1-model_v4	0.151	1096	633	0	237	1332	454	1199	1.733E-21	-2147483648
pgaptmp_006197	AF-P70388-F1-model_v4	0.108	514	429	0	16	529	430	911	2.586E+00	-2147483648
pgaptmp_006197	AF-Q54YH4-F1-model_v4	0.125	1377	939	0	259	1332	592	1968	3.866E-19	-2147483648
pgaptmp_006197	AF-Q54YZ9-F1-model_v4	0.369	1316	822	0	17	1332	546	1848	1.624E-63	-2147483648
pgaptmp_006197	AF-Q5DU28-F1-model_v4	0.095	470	386	0	159	585	958	1427	1.137E+00	-2147483648
pgaptmp_006197	AF-Q5AHA0-F1-model_v4	0.144	1316	1031	0	15	1330	1259	2464	3.586E-29	-2147483648

Command lines to reproduce:

> ${foldseek_path} version
941cd33ff0771cd2e3f144e3293e22a2b87e9fda

> $ echo ${foldseek_database}
/datasets/work/ev-agi-apps-db/reference/database/foldseek/foldseek_db_Alphafold_Swiss-Prot/foldseek_db_Alphafold_Swiss-Prot

> ${foldseek_path} easy-search pgaptmp_006197.pdb ${foldseek_database} pgaptmp_006197.foldseek.out ./tmp

pgaptmp_006197.pdb.gz

[milot-mirdita]

Interesting, this happens on the x86 server, but doesn't happen on Mac. I'll investigate. Thank you for the bug report

[milot-mirdita]

Oh, it does also happen, but with a 0 value instead of -INT_MAX

[milot-mirdita]

I pushed a fix. Thanks a lot for providing a test case. This made fixing the issue very straightforward!

[milot-mirdita]

By the way, do you have any idea why the structure you attached looks the way it does? The residues around 266 are very odd?