Integrating Knowledge Graphs into Machine Learning Models for Survival Prediction and Biomarker Discovery in Patients with Non–Small–Cell Lung Cancer

Survival prediction is a critical aspect of clinical study design and biomarker discovery. It is a highly complex task, given the large number of ″omics″ and clinical features, as well as the high degrees of freedom that drive patient survival. Prior knowledge can play a critical role in uncovering the complexity of a disease and understanding the driving factors affecting a patient′s survival. We introduce a methodology for incorporating prior knowledge into machine learning–based models for prediction of patient survival through knowledge graphs, demonstrating the advantage of such an approach for patients with non–small–cell lung cancer. Using data from patients treated with immuno-oncologic therapies in the POPLAR ([NCT01903993][1]) and OAK ([NCT02008227][2]) clinical trials, we found that the use of knowledge graphs yielded significantly improved hazard ratios, including in the POPLAR cohort, for models based on biomarker tumor mutation burden compared with those based on knowledge graphs. Use of a model-defined mutational 10-gene signature led to significant overall survival differentiation for both trials. We provide parameterized code for incorporating knowledge graphs into survival analyses for use by the wider scientific community. ### Competing Interest Statement All authors are employees of AstraZeneca at the time this work was performed and may have stock ownership, options, or interests in the company. [1]: /lookup/external-ref?link_type=CLINTRIALGOV&access_num=NCT01903993&atom=%2Fbiorxiv%2Fearly%2F2024%2F03%2F04%2F2024.02.29.582842.atom [2]: /lookup/external-ref?link_type=CLINTRIALGOV&access_num=NCT02008227&atom=%2Fbiorxiv%2Fearly%2F2024%2F03%2F04%2F2024.02.29.582842.atom