Search

OBJECTIVE: Clinical deep phenotyping and phenotype annotation play a critical role in both the diagnosis of patients with rare disorders as well as in building computationally-tractable knowledge in the rare disorders field. These processes rely on using ontology concepts, often from the Human Phenotype Ontology, in conjunction with a phenotype concept recognition task (supported usually by machine learning methods) to curate patient profiles or existing scientific literature. With the significant shift in the use of large language models (LLMs) for most NLP tasks, we examine the performance of the latest Generative Pre-trained Transformer (GPT) models underpinning ChatGPT as a foundation for the tasks of clinical phenotyping and phenotype annotation. MATERIALS AND METHODS: The experimental setup of the study included seven prompts of various levels of specificity, two GPT models (gpt-3.5-turbo and gpt-4.0) and two established gold standard corpora for phenotype recognition, one consisting of publication abstracts and the other clinical observations. RESULTS: The best run, using in-context learning, achieved 0.58 document-level F1 score on publication abstracts and 0.75 document-level F1 score on clinical observations, as well as a mention-level F1 score of 0.7, which surpasses the current best in class tool. Without in-context learning, however, performance is significantly below the existing approaches. CONCLUSION: Our experiments show that gpt-4.0 surpasses the state of the art performance if the task is constrained to a subset of the target ontology where there is prior knowledge of the terms that are expected to be matched. While the results are promising, the non-deterministic nature of the outcomes, the high cost and the lack of concordance between different runs using the same prompt and input make the use of these LLMs challenging for this particular task.

INTRODUCTION: New pharmaceuticals take on average 12 years to develop and receive FDA approval. However, there is abundant data in the public domain that can identify existing FDA-approved drugs that can be repurposed. Our goal was to develop graph-based ML and chemical structural-based ML models in a DoD-approved HIPAA compliant cloud environment to identify drugs and small molecules that can address biomedical impacts of high-altitude flight based on mRNA and miRNA signatures obtained from miniature swine high-altitude models. METHODS: We leveraged tooling to ingest, extract, and load databases into ARDIS cloud computing environment (AWS GovCloud IL4, HIPAA compliant, Defense Health Agency -- verified system). This tooling allows the efficient construction and remixing of knowledge graphs targeted for drug repurposing and application of modern machine learning algorithms to generate a ranked list of relevant small molecules. For this knowledge graph (KG), we semantically integrated data related to drugs, drug targets, microRNAs, protein pathways, and protein-protein interactions, along with several important biomedical ontologies that provide background biological information. We then investigated a graph machine learning model to predict drug -- miRNA interactions. We then used the SM2miR database to train a second ML model using miRNA data and extracted chemical structure from PubChem using CIDs. To visualize and provide explainability to the ML results, we developed a visualization component that displays nodes of interest and their relationships in the KG. RESULTS: In total, 36 links were predicted between drugs and any of differentially expressed pig genes and human homologs in high-altitude conditions. Broad classes of drugs with predicted links were statins/HMG-CoA reductase inhibitors, anti-cancer/neoplastic drugs, anti-infective drugs, and analgesic drugs. DISCUSSION: We created a knowledge graph environment and applied multiple ML models to identify drugs and small molecules that target differential gene expression associated with high altitude exposure. This platform can be applied to identify candidate marketed drugs for off-label use or repurposing for new indications. The next steps to validate drug predictions would be in vitro/in vivo studies with identified small molecules or retrospective analysis of clinical data sets to confirm therapeutic effects. Learning Objectives 1. The audience will learn about differential gene expression found in an animal model of high altitude flight. 2. The audience will learn about machine learning algorithms that can be applied to drug repurposing. 3. The audience will learn predicted links between classes of drugs and differential gene expression found in high altitude exposure. [ABSTRACT FROM AUTHOR]