(4bh) Next-Gen Biosynthesis Planning

Research Interests: (Retro)-biosynthetic approaches in synthetic biology have made significant advances in designing (i) production routes for new biochemicals, (ii) novel biodegradation strategies of toxic compounds, and (ii) funneling of heterogeneous aromatics towards bio-renewable chemicals with industrial interest. Pathway design is an integrated task that requires knowledge and encoding of relevant biochemistries, a biophysical understanding of enzymatic activity and specificity, an understanding of the thermodynamic feasibility of individual steps, and subsequent host selection and metabolic engineering. However, using only cataloged enzymatic activities can limit the discovery of novel production routes. Through the use of retro-biosynthesis algorithms, novel conversions can be assembled that leverage alteration of enzymatic substrate or cofactor specificity. These novel pathways may shorten existing routes, and avoid toxic intermediates or low-capacity conversions. The need to modify existing enzymes gives rise to enzyme engineering challenges in implementing these designs. For that, we developed novoStoic2.0, an integrated workflow for pathway synthesis that combines different tools for de novo pathway design within a single framework. dGPredictor estimates Gibbs energy change for (novel) reactions involving possibly novel structures to assess the thermodynamic feasibility of various steps. The integration of dGPredictor within novoStoic2.0 safeguards against using reactions (or novel conversion steps) in a thermodynamically unfavorable direction. EnzRank is a convolutional neural network (CNN) based approach to select enzyme candidates for novel conversions in pathway design. EnzRank rank-orders existing enzymes for their suitability to undergo successful protein engineering through directed evolution or de novo design towards desired specific substrate activity. EnzRank supplements pathway designs provided by novoStoic to find enzyme candidates for any novel reaction step. novoStoic2.0 provides a unified web-based interface for the biosynthesis of thermodynamically feasible, carbon and energy-balanced pathways with a way of selecting enzymes to undergo re-engineering for novel reaction steps.

With advancements in high-throughput experimentation and AI algorithms, there is an opportunity to develop closed-loop Design-Build-Test-Learn (DBTL) systems for biosynthesis. These systems can generate high-fidelity data that enhance synthesis planning tools in conjunction with high-throughput experimentation. This approach will enable the synthesis of a wide range of chemicals using enzymes for applications such as drugs, biofuels, bioplastics, and value-added chemicals. My goal is to develop AI-based agents capable of designing synthesis pathways, utilizing large language models to mine the literature and identify experimental procedures. These agents will collaborate with high-throughput automated systems to conduct experiments and learn from the outcomes, thereby continuously improving the entire synthesis process in future iterations to develop new tools for Next-Gen chemical bio-syntesis planning.

Teaching Interests: I am interested in teaching Machine Learning and Artificial Intelligence (ML/AI) and optimization algorithms, with a focus on their applications in chemical synthesis and chemical engineering. Specifically, I am passionate about integrating ML/AI with metabolic engineering and synthetic biology to develop new algorithms and innovative solutions. My goal is to help students understand the potential of these technologies in designing efficient pathways for chemical production. I am eager to teach foundational concepts, guide hands-on projects, and inspire students to explore the intersection of AI and chemical engineering, preparing them for future challenges in the field.