Appears in collection : Interplay between AI and mathematical modelling in the post-structural genomics era / Interaction entre l'IA et la modélisation mathématique à l'ère post-génomique structurale

I will give an update on Foldseek, which can perform structural similarity searches and alignment through the Uniprot50 and MGnify/ESM30 database of 95M structures in a few seconds on a single core, at a sensitivity similar to TMalign (search.foldseek.com). At its core is the 3Di structural alphabet that we learned using a discrete variational autoencoder. I will also present SpaceDust, a tool for fast sequence- and structure-based searches of clusters of 'positional orthologs' between genomes, as well as PenguiN, our new strain-resolved viral metagenome assembler.