Implications and lessons learned from foundational AI-models applied to genomic/proteomic data

We present our recent experience in developing large-language models for biological applications, typically examined from the lens of their ability to capture complex dependencies that occur within genomic and proteomic datasets. We comment on the appropriateness of these models in driving novel biological expeirments, especially from the point of view of designing novel antimicrobial peptides or in discovering new variants of enzymes.