MIT chemical engineers published a study introducing an AI-driven method that programs industrial yeast to produce biopharmaceutical proteins faster and at lower cost. The team trained large language models on yeast genomic and regulatory sequences to predict DNA edits that improve expression and secretion of target proteins. The approach aims to shorten strain-engineering timelines and reduce reliance on iterative wet-lab screening. The research was led at MIT and validated on multiple industrial yeast strains and expression systems. By automating the design of regulatory sequences and production pathways, the method could compress early process development and improve yield for vaccines and biologics. Large language models (LLMs) are statistical models trained on extensive sequence or textual data to predict likely next elements—here repurposed to propose genetic edits for yeast optimization. A separate study echoed the same concept: modeling yeast DNA “language” to tune expression and scale protein output. Together, the papers show converging industry interest in AI-first strain design, signaling faster adoption of generative models in biologics manufacturing workflows.