Igor Tolstoy

[Igor Tolstoy @vircurator]

@tangming2005 sra, genbank, geo-archieval databases, submitter’s centric, only submitter can change metadata.
but they are not really care about it after publication.
new layer where data can be curated needed.
I know some viruses can be reclassified, just no place to save this information

[Igor Tolstoy @vircurator]

@milja001 @TSPKU @MarthaClokie @panagouras3 @bentpetersen NCBI aware about this issue for many years, they just do not care. Same way ignored all ICTV proviral and retroviral species.

[Stephen Turner 🦋 @stephenturner.us @strnr]

Biological databases in the age of generative artificial intelligence https://t.co/3NhG11r0oK 🧬🖥️🧪

[Igor Tolstoy @vircurator]

@milja001 @remidenise @TSPKU you can draw it as baselines for each year

[Igor Tolstoy @vircurator]

@milja001 @remidenise @TSPKU Can you add ICTV VMR phage genomes count on your web page?
it looks like good replacement for NCBI phage refseq count.

https://t.co/YP3LIXsrDH

[Igor Tolstoy @vircurator]

https://t.co/2aYGSBowWV

[Igor Tolstoy @vircurator]

@BlackTmiR sure, I have made copy of major species germlines just four days ago : between shutdowns.

https://t.co/yZO8cNCY18
time-outed now.

[Igor Tolstoy @vircurator]

https://t.co/Hi58hNoZOZ down second time in a week, two days in a row.
What is going on?

[Igor Tolstoy @vircurator]

@BlackTmiR github:
nextflow/airrflow broken
ANARCI installation process broken
what else?

[Igor Tolstoy @vircurator]

@StevenSalzberg1 https://t.co/DxhobocHFD

[Igor Tolstoy @vircurator]

@StevenSalzberg1 @markusjsommer Do you have version of Balrog with the same functionality? Just evaluate quality of prokaryotic proteins.Interesting problem: identify incorrectly annotated N-terminus and C-terminus due intersections with https://t.co/KUb4MpGsEK I understand, your TCN can assign score to each AA

[Igor Tolstoy @vircurator]

https://t.co/frKbpkuODi

~700K predicted structures,
future structural genbank?

it really need to be indexed with foldseek

[Igor Tolstoy @vircurator]

@thesteinegger Have you tried validate GTDB reference core MSAs with 3di+foldmason?

[Igor Tolstoy @vircurator]

@duguyuan @thesteinegger The most questionable trick in 3Di model.

During my experiments with 3Di training
I modified code and predicted Xn from Xn (identity mode) , VAE-VQ allow this hack.
I would say, on validation set FP1 was just slightly smaller than original.

[Igor Tolstoy @vircurator]

@duguyuan @thesteinegger Just, curious
Is it possible train model which
will distinguish methods how structure was folded?
let’s say af2,esm2,pdb, may be af3,etc…

[Igor Tolstoy @vircurator]

@ViralZone it looks like publications which use for structure analysis foldmason or muscle-3d will be published only next year.
it will be very interesting to see which one is better on real data.

[Igor Tolstoy @vircurator]

@duguyuan @thesteinegger I know , AF 3Di is slightly better, but much
slower to generate.
Some time ago we discussed with you possibility predict AF2 3Di from ESM2 3Di.
My point -3Di is not final choice ,
for downstream tasks better linear alphabet can be invented.
for example
https://t.co/bi1Tx1nAj3

[Igor Tolstoy @vircurator]

@duguyuan @thesteinegger 3Di just good structures indexing trick .Do not make silver bullet from it. From the same protein sequence AF2 and ESM2 produce 3Di which are different in average in 15% positions(2M proteins).Take these two 3Dis and predict AA sequences, how different they will be in average?

[Igor Tolstoy @vircurator]

@sdeorowicz Have you tried compress >10millions covid genomes (length 30k) ? interesting how much space and time it will take? Can you provide some estimations?