Emily Kurtz
Pronouns:
Research Mentor(s): Sriram Chandrasekaran
Research Mentor School/College/Department: Biomedical Engineering / Medicine
Program:
Authors: Emily Kurtz, Ryan Schildcrout, Sriram Chandrasekaran
Session: Session 2: 10:00 am – 10:50 am
Poster: 16
Abstract
Metabolism is an incredibly complex process with many different interactions. Our goal is to determine how metabolism is regulated at the transcriptional, post-transcriptional, translational, and post-translational levels with different omics data. To accomplish this, we aim to generate predictive models using machine learning algorithms. In particular, we have been using ridge regression and lasso regression models. Both of these models are a least-squares regression line with a penalty depending on the weight of the data on our trendline. We are taking data obtained from a public database, the Cancer Cell Line Encyclopedia, and using most of it to train the machine learning algorithm, and predicting the result using the remainder of the dataset. We do this multiple times and average the results using cross validation. We then compare the predicted results with the true results and analyze how good of a predictive model each type of omics data is with correlations and Bonferroni-corrected p-values. While the research is currently ongoing, we have managed to generate ridge regression models for six types of omics data. We have found that ridge regression is able to predict metabolites using phosphoproteomics accurately 56.9% of the time, histone post-translational modifications (PTMs) 68.9% of the time, proteomics 75.1% of the time, genomics 84.4% of the time, miRNA 91.1% of the time, and transcriptomics 98.2% of the time. This indicates that transcriptomics has more influence over cellular metabolism than histone PTMs and phosphoproteomics. We hope to continue finding correlations between the accuracy of machine learning and influence on metabolism using different omics data. With this information, not only will we be able to predict metabolic data but this data will help us in understanding cancer metabolism.
