Abstract 3531: PRESSNET: Patient stratification and biomarker discovery using multi-modal knowledge graph framework

Abstract Background: Multiomics data is critical to obtain a near comprehensive picture of disease progression and drug response. In addition, the generation of response and survival biomarkers, and the segmentation of patients into subtypes with distinct, actionable ‘omic signatures and survival trajectories, is vital for personalised medicine research and successful trial design. However, as the volume and diversity of data increases, so too does the challenge of effective multiomic data integration. Knowledge graphs (KGs) can capture heterogeneous data and relationships between entities in a flexible and scalable data structure, making them suitable for this domain. We have developed PRESSnet, a framework for building Patient KGs and analysing clinical data for novel patient stratification hypotheses and clinical biomarker discovery. Methods: PRESSnet is an end-to-end framework that turns raw patient data into graph-derived insights. Firstly, the user chooses which modality files to include, what assumptions to make about data processing, and which graph algorithms to use. PRESSnet then automatedly creates a patient KG of the input data, where nodes represent patients and their associated features. In addition, biomedical prior knowledge, for example in the form of gene-pathway or gene-gene relationship data, is also integrated with the graphs. Insights are generated from the KG via community detection, graph embedding generation, and graph neural networks; these generate hypotheses for novel patient subtypes or biomarkers for clinical outcomes. As graph algorithms capture interrelationships between nodes, PRESSnet offers biomarkers that are composite, i.e. that can contain features from multiple ‘omics and clinical features. Results: We applied PRESSnet to the MSK 20221 cohort of IO-treated LUAD patients, and it uncovered prognostic composite biomarkers that stratified biomarker-positive patients from the whole cohort with a p value < 0.001 (95% CI) for OS, including known markers of poor prognosis such as STK11, RBM10, KRAS and KEAP1 mutations, and high neutrophil/lymphocyte ratio. We also generated embeddings of patients in the cohort and predicted survived/deceased status with an AUC of 0.82, outperforming published state-of-the-art. Conclusions: We have successfully developed a generalisable framework for generating insights from patient data using state-of-the art knowledge graph data science. PRESSnet can generate novel stratification and biomarker hypotheses that can potentially inform the next generation of IO targets and clinical biomarkers. Footnotes1 Vanguri et al., “Multimodal integration of radiology, pathology and genomics for prediction of response to PD-(L)1 blockade in patients with non-small cell lung cancer”, Nat Cancer. 2022 Citation Format: Krishna C. Bulusu, Jake Cohen-Setton, Ioannis Kagiampakis, Miguel Goncalves, Gavin Edwards, Sanddhya Jayabalan, Shruti Shikare, Kelvin Tsang, Ben Sidders, Etai Jacob. PRESSNET: Patient stratification and biomarker discovery using multi-modal knowledge graph framework [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 3531.