Computational Analysis of Physiological Systems at Multiple Time and Length Scales

Abstract

Advancements in unsupervised and supervised machine learning algorithms have provided new insights into data patterns and generated prediction models for practical usage. These machine learning approaches have been evolving our understanding of various physiological and biological systems. This is especially true in the era of big data - how to leverage large-scale multi-source information to model physiological phenomena and interpret observations from an unprecedented computational perspective. In this dissertation, I focus on three projects related to different physiological systems at multiple time and length scales. First, I describe a network approach for comparing and contrasting structural dynamics of evolutionarily-related protein families through Principal Component Analysis (PCA) and Molecular Dynamics (MD) simulation at the atom level. Next, I describe a trans-tissue method for predicting proteomics from transcriptomics in cancers at the tissue level, using a classical machine learning model. Finally, I describe a deep convolutional neural network model for automatic segmentation of sleep arousals based on human polysomnograms at the organism level. These novel computational approaches will potentially facilitate future research in the fields of conformational comparison of proteins, proteogenomics, and signal processing of sleep recordings.