USE OF ARTIFICIAL INTELLIGENCE IN QUANTITATIVE PROTEOMIC DATA ANALYSIS

Abstract

In the Langlais Lab, we are investigating novel proteins involved in insulin-stimulated glucose uptake (ISGU) to discover key mechanistic proteins that were unknown before. The goal of this project was to see if artificial intelligence could help aid in the discovery of these proteins. Quantitative proteomics comparing 3T3-L1 fibroblasts and differentiated adipocytes identified over 7,800 proteins, which were refined to 914 candidates through general screening. Manual sorting through 914 proteins to find potential candidates and perform experiments on them proved to be a timely task. However, with the help of LLM's such as ChatGPT it was hypothesized that they could drastically narrow the list down to more relevant proteins. Through stringent testing of many different AI models, the ChatGPT Deep Research model was found to accurately sort through proteins and identify key mechanistic ones involved in ISGU. It also provided hypotheses on why these proteins warrant further investigation. ChatGPT Deep Research met key performance benchmarks in prompt comprehension, biological insight, data processing, visual representation, and accuracy.