Biologically informed deep neural network for prostate cancer discovery

• Published in: Nature (London) Vol. 598; no. 7880; pp. 348 - 352
• Main Authors: Elmarakeby, Haitham A., Hwang, Justin, Arafeh, Rand, Crowdis, Jett, Gang, Sydney, Liu, David, AlDubayan, Saud H., Salari, Keyan, Kregel, Steven, Richter, Camden, Arnoff, Taylor E., Park, Jihye, Hahn, William C., Van Allen, Eliezer M.
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 14.10.2021; Nature Publishing Group
• Subjects: 119/118; 13; 13/1; 631/67/69; 639/705/117; Artificial neural networks; Cell cycle; Cell Cycle Proteins - genetics; Cell growth; Deep Learning; Diagnosis; Drug Resistance, Neoplasm - drug effects; Drug Resistance, Neoplasm - genetics; Fibroblast growth factor receptor 1; Genomics; Health aspects; Humanities and Social Sciences; Humans; Learning algorithms; Machine learning; Male; Metastasis; Modelling; multidisciplinary; Mutation; Neural networks; Patients; Phenotypes; Prostate cancer; Prostatic Neoplasms - diagnosis; Prostatic Neoplasms - drug therapy; Prostatic Neoplasms - genetics; Proto-Oncogene Proteins - genetics; Receptor, Fibroblast Growth Factor, Type 1 - genetics; Receptors, Androgen - genetics; Reproducibility of Results; Science; Science (multidisciplinary); Therapeutic targets; Treatment resistance; Tumor Suppressor Protein p53 - genetics; Tumors; United States
• ISSN: 0028-0836; 1476-4687; 1476-4687
• DOI: 10.1038/s41586-021-03922-4

The determination of molecular features that mediate clinically aggressive phenotypes in prostate cancer remains a major biological and clinical challenge 1 , 2 . Recent advances in interpretability of machine learning models as applied to biomedical problems may enable discovery and prediction in clinical cancer genomics 3 – 5 . Here we developed P-NET—a biologically informed deep learning model—to stratify patients with prostate cancer by treatment-resistance state and evaluate molecular drivers of treatment resistance for therapeutic targeting through complete model interpretability. We demonstrate that P-NET can predict cancer state using molecular data with a performance that is superior to other modelling approaches. Moreover, the biological interpretability within P-NET revealed established and novel molecularly altered candidates, such as MDM4 and FGFR1 , which were implicated in predicting advanced disease and validated in vitro. Broadly, biologically informed fully interpretable neural networks enable preclinical discovery and clinical prediction in prostate cancer and may have general applicability across cancer types. A biologically informed, interpretable deep learning model has been developed to evaluate molecular drivers of resistance to cancer treatment, predict clinical outcomes and guide hypotheses on disease progression.