Predicting Psychiatric Hospital Admission Among Adults with Major Depressive Disorder / Predicting drug sensitivity of tumor cell lines from genomic data using deep learning

Seminar Date: 
2016-04-08
Seminar Time: 
11am - 12pm
Seminar Location: 
5607 Baum Boulevard, Room 407A
Presenter: 
Sergio Castro and Michael Ding

Predicting Psychiatric Hospital Admission Among Adults with Major Depressive Disorder

Abstract: Hospital admission of adults with major depressive disorder is associated with higher levels of disability and healthcare costs. Early identification of patients at risk of admission can help reduce unfortunate outcomes, and tailor treatment strategies to prevent future hospitalizations. The aim of this study is to develop a model for prediction of psychiatric hospital admission 6 months after a psychiatric visit. Here we describe a model for prediction of psychiatric hospital admission 6 months after a psychiatric visit using data from a Colombian database. Our evaluation showed that it is possible to predict future Major Depression admissions with moderate accuracy using data from multiple sources. The use of computational models in psychiatry could provide inmense value due to the lack of biomarkers for risk stratification on specific outcomes.

Predicting drug sensitivity of tumor cell lines from genomic data using deep learning 

Recent large scale pharmacogenomics studies have collected information on gene expression levels, copy number alterations, genetic mutations, and drug sensitivities of a large number of tumor cell lines. These studies have enabled the development of computational methods for predicting drug sensitivity of tumors from genomic data. Unfortunately, the high dimensionality of these datasets hinders meaningful analysis. We use a deep autoencoder to generate low dimensional representations of data from the Cancer Genome Project pharmacogenomics study. Using these representations as feature sets allows the generation of drug sensitivity models with good predictive accuracy. This demonstrates potential for the application of deep learning for dimensionality reduction of biomedical datasets.

 

^