Title: Programmable Protein Therapeutics via Generative Language Models

Abstract:

CRISPR has revolutionized biotechnology by enabling the simple design of guide RNAs to target and edit almost any DNA sequence. By developing new generative protein design algorithms, my hybrid lab focuses on extending this CRISPR-like programmability to proteins and other key molecules. In this talk, we will first delve into our algorithms that design binders to undruggable proteins, such as those driving pediatric cancers (alveolar rhabdomyosarcoma and Ewing’s sarcoma) and neurodegenerative diseases (Huntington’s and Alexander Disease). Our generative language models, including SaLT&PepPr, PepPrCLIP, and PepMLM, design short binding peptides from target sequence alone, with no dependence on stable 3D structures, and by fusing these "guide" peptides to E3 ubiquitin ligases, deubiquitinases, and other modifying enzymes, we have created a CRISPR-analogous system to edit these proteins. To be even more specific, we train isoform-specific targeting models such as PTM-Mamba for PTM-specific binding, FusOn-pLM for fusion oncoprotein-specific degradation, and moPPIt for motif-specific targeting of protein-protein interactions. Inspired by the power of language models, we further show how we can extend this programmability to DNA with our PAM-free CRISPR enzymes and our recent DPAC model, as well as heavy metals through our MetaLATTE algorithm and chemical pollutants, such as PFAS. Finally, we will explore our long-term goal of generating new cell states with model-designed proteins, highlighting our recent work on transcription factor-directed stem cell differentiation to ovarian cell types, such as granulosa cells and oogonia. By combining generative design with experimental engineering, our hybrid lab aims to translate these advances into practical applications for treating intractable diseases and addressing environmental challenges.