Generic Multi task Learning Framework for Structural Mixed Data using Deep Neural Network Feature Sharing Architecture Application to Predictive Tasks in Vietnamese Herbal Medicine
Abstract
Both traditional and deep neural network machine learning techniques have been shown effective to solve difficult problems in computational biology such as ligand based virtual screening, reactivity to biologival macromolecules, etc. Recently Multi task Learning (MLT) has emerged as a powerful tool for computational drug discovery. MLT is a subfield of machine learning in which multiple learning tasks are solved at the same time by exploiting commonalities and differences across tasks. In this proposal, we plan to develop a generic multi task learning method which is especially tailored to data that have internal strucures and are of mixed types using feature sharing mechanism of deep neural network. We choose a domain of Vietnamese herbal medicine, and verify the proposed method by constructing a web based component to solve Quantitative Structure Activity Relationship to predict different aspects of drug like molecule-metabolites such as anti cancer activity, physical chemistry and physical drug properties. To evaluate the MLT model, we compare it to other popular single task learning models used in pharmacological field. From these results, we expect to build a biological attributed network for each Vietnamese herbal medicine which contains metabolites with their potential characterization in treatment product. The potential result could be used to build a component which is integrated into Vietnamese Herbal Medicine Database website. Further, the developed framework could be used as a generic tool to solve various predictive tasks simultaneously for structural mixed data.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 14, 2022
- Source ID
- FA23861914032XX0
Entities
People
- Le Ly
Organizations
- Air Force Office of Scientific Research
- International University
- United States Air Force