The genome says what a cell could make; the proteome says what it’s actually doing — and it’s the hardest omics layer to read, because there’s no PCR for proteins. Deep learning is cracking it: transformer models now sequence peptides straight from a mass spectrum with no database, surfacing peptides and even organisms no database contained. The frontier is pushing this to single cells and into foundation models — and a brand-new Nature Methods review draws the line from protein identification all the way to AI virtual cells.
