149 Proteomics Analysis of FLT3-ITD Mutation in Acute Myeloid Leukemia Using Deep Learning Neural Network. (11th January 2018)

• Record Type: Journal Article
• Authors: Liang, Christine; Wahed, Amer; Nguyen, Nghia Andy
• Is Part Of: American journal of clinical pathology. Volume 149(2018)Supplement 1
• Page Start: S64
• Page End: S64
• Publication Date: 2018-01-11
• Subjects: Diagnosis, Laboratory -- Periodicals; Pathology -- Periodicals; 616.07
• DOI: 10.1093/ajcp/aqx121.148
• Languages: English
• ISSNs: 0002-9173

Abstract

Abstract: Big data analytics in cancer proteomics and genomics, where raw data are largely uncategorized, can significantly benefit from Deep Learning. Deep Learning algorithms extract high-level, complex abstractions as data representations through a hierarchical learning process. In the present study, we explore how Deep Learning can be utilized for proteomics analysis in acute myeloid leukemia (AML). Specifically, we attempt to determine a set of critical proteins that are associated with the FLT3-ITD mutation out of 231 proteins available in newly diagnosed AML patients. We implement a Deep Learning network consisting of autoencoders that are stacked to form hierarchical deep models from which high-level features are compressed, organized, and extracted, without labeled training data. Dimensional reduction is initially performed with supervised training to reduce the number of critical proteins from 231 to 20. The initial use of the full attribute set of 231 proteins yields 72% accuracy for the conventional network. The Deep Learning network performs at 81% accuracy. Using the top 20 ranked proteins in this initial trial, the best accuracy is obtained by the Deep Learning network at 97%. Deep Learning with unsupervised training found an excellent correlation between the FLT3-ITD mutation and levels of these 20 proteins (sensitivity of 90%, specificity of 100%). Proteomics data show that there are 231 proteins associated with AML patients with FLT3-ITD as their sole mutation. By using a Deep Learning network, we were able to hone in on the 20 proteins with the strongest association with FLT3-ITD. The results of this study allows a more focused approach to determining critical protein pathways in the FLT3-ITD mutation, greater effectiveness in monitoring chemotherapy response, and a more personalized treatment protocol. This study provides proof of concept for a more accurate approach in modeling big data in cancer proteomics and genomics.