AI-powered Onco-Knowledge Graph (OncoKG) for Precision Medicine

Clinical knowledge graphs have profound usage in biomedical research due to their ability to integrate and structure enormous amount of heterogenous biological information scattered across various databases. These graphs can serve as prime resources for querying gene-gene, gene-disease, drug-disease and drug-target associations as well as kinase-substrate interactions. Drug discovery being expensive and time-consuming, these knowledge bases can be leveraged to identify novel interactions that enhance decision-making in drug repurposing and precision medicine. However, to make maximum utilization of such a knowledge repository, queries related to drugs or drug targets need to be precise and focused. Consequently, knowledge graphs can be utilized to provide evidence-based recommendations for tailoring treatment in patients based on their unique molecular profile.

In oncology, standard frontline chemotherapy is often not curative, and the disease frequently relapses. Relapse, attributed to drug resistance and refractoriness, severely limits the patient's chances of being cured. Patients often exhaust therapy options, leaving fewer effective therapies available. Despite the increasing availability of next generation sequencing and omics information for these patients, identifying actionable and highly effective targets remain challenging. This underscores the unmet need of a model capable of identifying molecular targets and rank potent drugs in order of their efficacy for a more effective personalized treatment.

To address this gap, we introduce an AI-powered framework with active learning designed to predict drugs for precision medicine based on patients" omics profiles. This framework includes an Onco-Knowledge Graph (OncoKG) and an AI module to identify novel drug targets from multi-omics data along with a Large Language Model (LLM)-enabled interface to query the OncoKG. To assess the tool, we utilized publicly available multi-omics datasets of Pancreatic Ductal Carcinoma (PDAC) from Clinical Proteomic Tumor Analysis Consortium (CPTAC) and Acute Myeloid Leukemia (AML). Preliminary results validate targets identified for PDAC and AML to be clinically relevant. Importantly, drugs suggested by the AI-OncoKG framework show promise and warrant further investigation invitro and invivo. With further pre-clinical examinations, this tool has the potential to revolutionize personalized treatment with an outlook to facilitate drug choices and repurposing.