Personalized therapy requires innovative genome-scale studies for identifying expression patterns in disease progression. Since each individual feature selection algorithm has different strengths, hybrid models, combining multiple algorithms, have become necessary for identifying clinically relevant biomarkers. Furthermore, it is important to reveal disease-mediated biomarker interactions, including feedback circuits, for more effective therapy. This proposal will develop a novel bioinformatics framework by combining genomics, proteomics, and clinical approaches for more informed clinical decision-making.
In Aim 1, we will develop a feature selection system by integrating multiple algorithms for biomarker identification. Combinations of several feature selection methods in different stages of gene filtering will be investigated. The optimal combination scheme for generating the highest prediction accuracy with the minimum number of biomarkers will be determined for several cancer types. The identified biomarkers will be validated by extensive public data sets.
In Aim 2, we will develop a novel methodology for modeling biomarker interaction patterns for clinical classification. Based on the expression profiles of the biomarkers selected in Aim 1, Dempster-Shafer belief networks will be employed for predicting individual clinical outcome. The network structure will elucidate molecular interactions among the biomarker proteins in disease progression. Algorithms will be developed to optimize the performance of the Dempster-Shafer network formalism. Different combination rules of Dempster-Shafer theory will be implemented in the belief networks to handle various real- life clinical applications.
In Aim 3, stringent criteria will be applied to compare Dempster-Shafer networks with Bayesian networks and other machine learning methods, using the same data sets. The best molecular classifiers will be identified and evaluated with respect to traditional prognostic factors. This strategy will allow patient stratification based on risk of tumor recurrence and the need for adjuvant chemotherapy. The biomarker interactions derived in Dempster-Shafer networks and Bayesian networks will be evaluated for providing useful biological insights. A web-based infrastructure for clinical decision-making will be developed and validated. This project will focus on predicting metastasis and relapse in non-small cell lung cancer and colorectal cancer. This multidisciplinary research will involve collaborations among bioinformaticians, clinicians, and biomedical researchers for algorithm development and evaluation with respect to strategies for biomarker- based patient stratification and assessment of therapeutic outcomes in different prognostic groups. The project results will be evaluated in prospective clinical trials for colorectal cancer treatment intervention. Our long-term goals are to identify biomarkers that reveal important molecular mechanisms and/or therapeutic targets underlying disease and to make accurate clinical predictions for personalized therapy. This study will advance the computational modeling of human genome data in disease for clinical decision-making.
Local and distant recurrence is the major cause of cancer-related deaths. This research will develop an innovative computational system with implications for early detection of cancer as well as potential recurrence and metastasis, leading to increased survival rates. This study will advance biomedical informatics for applications in clinical decision- making.
Showing the most recent 10 out of 30 publications
