Research Overview
Osteosarcoma (OS) is the most common primary bone malignancy and most genomically complex cancer. The 5-year survival remains unchanged in the past 30 years. Early occurrence of pulmonary metastasis is the main challenge for patients to be cured. Due to its high heterogeneity, prognosis of osteosarcoma diverse greatly and no clinical evidence could be applied to predict patient outcome. The fact that lacking of prognostic marker for osteosarcoma hampers precision medical care for individual patient. More importantly, early detection and phenotyping of OS would enhance the chances to cure.
Previous studies have shown the potential use of a metabolomics in distinguishing OS’s metabolomics profile compared with benign bone tumor and healthy controls. Osteosarcoma metabolomics showed much more coherency than genomics, which make it as promising candidate for predictive markers. However, due to the significant variation across the covariances of these metabolites, the design of a predictive model is still challenging.
This project will design, test and validate a machine learning model to predict OS from metabolites using a non-convex kernel based algorithm that can address this challenge.
The project aims at establishing a prognostic predictive model using machine learning in osteosarcoma metabolomics. This model will be potentially useful for providing adjuvant information to stratify osteosarcoma patients and subsequently to guide clinical decision.
Research Team
Kayvan Najarian, Ph.D., PI, University of Michigan, Data Scientist, Department of Computational Medicine and Bioinformatics, Director of Data Science at MichiganCenter for Integrative Research in Critical Care (MCIRCC).
Yingqui Hua, M.D., Ph.D., PI, Shanghai Jiao Tong University (SJTU), Orthopedic Oncologist, Department of Orthopedics, Shanghai General Hospital.