+441518081136Heuristic multi-objective optimization algorithm to extract biomarker based on mutation combinations in whole gene information for disease diagnosis
Annual Congress on Oncology and Biomarkers Summit
November 27-28, 2017 | Atlanta, USA
Yong-Joon Song
Korea Advanced Institute for Science and Technology, South Korea
Posters & Accepted Abstracts : J Med Oncl Ther
Abstract:
There are still many efforts to diagnose diseases early. Among them, molecular diagnostics using biomarker is one of the powerful tools that can pre-diagnose diseases before symptoms appear. In critical diseases such as acute myeloid leukemia (AML), when the symptoms are present, it is already late. It is possible to bring about complete cure of cancer by performing preliminary examination and early treatment based on molecular diagnostics. However, researches related to biomarkers have been done only from a biomedical point of view, focused on specific gene sequences or protein expressions that are thought to be related to disease. To overcome this stereotype, we proposed an algorithm that uses a combination of disease-related mutation information from entire gene. We used NGS data of solid tissue normal samples from skin and primary solid tumor samples from bone marrow, which were obtained from 50 AML patients from TCGA database. In addition, we extracted mutation information by using GATK tool. In order to extract only cancer-related mutation information among the obtained mutation information, we use a following proposed algorithm. There are millions of mutations in the entire gene, and a huge number of combinations. Thus, in order to find biomarker with low complexity, we sorted all the mutations by scoring how well the disease and normal samples could be separated by each mutation. In this process, the case of genetic mutation that occurs due to the difference of skin and bone marrow was excluded. In the derived list of mutations, we obtained optimal biomarkers by heuristically solving three multi-objectives optimization problems, which includes three parameters such as disease classification ratio, the distance of inter-clusters and the distance of intra-clusters. Using proposed method, we could get a mutation combination that has 100% disease classification performance for the sample we acquired.
Biography:
Yong-Joon Song has completed his bachelor’s degree in Electrical Engineering at KAIST in 2016. Currently, he is a PhD student in school of electrical engineering at KAIST. His research interest area is Bioinformatics.
PDF HTML