Democratizing Multi-Omics to Expedite Discovery of Hidden Metabolic Pathways

PROJECT SUMMARY/ABSTRACT There is a fundamental gap in our understanding of how metabolism changes in many diseases because we lack methods for high-throughput, unbiased discovery of indirect metabolite-protein connections. Continued ex- istence of this knowledge gap represents a major issue for public health and the mission of the NIH because, until it is filled, development of treatments for many diseases will remain largely intractable. Multi-omic analysis of proteomes and metabolomes from the same system offers a promising path to discover hidden metabolic pathways, but the requirement for human expert interpretation is a critical barrier that prevents complete value extraction from multi-omic experiments. The long-term goal of the Meyer Research Group at Medical College of Wisconsin is to reveal previously hidden metabolic pathways. The overall objective here, which is the first step in realizing this vision, is to democratize multi-omic data collection and data interpretation, thereby increasing the pace of metabolic pathway discovery. The central hypothesis is that artificial intelligence models can learn to draw new metabolic connections between metabolites and proteins. This hypothesis is based on preliminary data generated by the applicant and published literature, which shows how the strategy reveals known and new connections between metabolites and proteins. The rationale for the proposed research is that unbiased, data- driven discovery of new metabolic connections with AI algorithms (such as deep neural networks) will result in new and innovative therapeutic targets that can be manipulated positively or negatively to prevent or treat dis- ease. Guided by preliminary data and literature, this hypothesis will be tested by pursuing two complementary focus areas: (1) multi-omic data integration, and (2) multi-omic data collection. The multi-omic data integration focus uses AI models, already established as feasible in the applicant’s lab, to predict metabolite-protein inter- actions. AI models will be optimized with existing public data, models will be validated with newly collected data, and then novel metabolic connections will be validated using classic genetic and biochemical techniques. The second focus area builds new, fast methods for multi-omic data collection to feed data into AI models, starting from a recent advancement published by the applicant (Meyer et al., ChemRxiv 2020, accepted at Nature Meth- ods). The applicant’s lab will further develop this method to quantify the full yeast proteome, and also extend the method to enable multi-omic analysis on a single platform. This approach is innovative because it departs from the status quo of slow multi-omic data interpretation requiring expert humans by building and validating a new, automated AI method for metabolite pathway discovery. The multi-omic data collection focus is innovative be- cause it departs from the status quo of slow multi-omic data collection requiring multiple platforms and hours per sample by enabling unified multi-omic analysis in minutes. This contribution will be significant because ulti- mately, the knowledge, validated methods, and resource datasets generated by this project will open new hori- zons in drug development for diseases with altered metabolism, such as cancers and diabetes.