Controllable protein design with language models

The twenty-first century is presenting humankind with unprecedented environmental and medical challenges. ability to design novel proteins tailored for specific purposes would potentially transform our respond these issues in a timely manner. Recent advances the field of artificial intelligence are now setting stage make this goal achievable. Protein sequences inherently similar natural languages: amino acids arrange multitude combinations form structures that carry function, same way as letters words sentences meaning. Accordingly, it not surprising that, throughout history language processing (NLP), many its techniques have been applied protein research problems. In past few years we witnessed revolutionary breakthroughs NLP. implementation transformer pre-trained models has enabled text generation human-like capabilities, including texts properties such style or subject. Motivated by considerable success NLP tasks, expect dedicated transformers dominate custom sequence near future. Fine-tuning on families will enable extension their repertoires could be highly divergent but still functional. combination control tags cellular compartment function further controllable functions. Moreover, recent model interpretability methods allow us open ‘black box’ thus enhance understanding folding principles. Early initiatives show enormous potential generative functional sequences. We believe using create promising largely unexplored field, discuss foreseeable impact design. Both essentially based sequential code, feature complex interactions at multiple scales, which can useful when transferring machine learning from one domain another. Review, Ferruz Höcker summarize models, transformers, application