Young, Jonathan
(2020)
DEEP LEARNING FOR CAUSAL STRUCTURE LEARNING APPLIED TO CANCER PATHWAY DISCOVERY.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
This is the latest version of this item.
Abstract
It is well appreciated that cancer is a disease of aberrant signaling and the state of a cancer cell can be described in terms of abnormally functioning cellular signaling pathways. Identifying all of the abnormal cellular signaling pathways causing a patient’s cancer would enable more patient-specific and effective treatments. Here we interpret the cellular signaling system as a causal graphical model and apply a modified deep neural network to learn latent causal structure that represents cancer pathways. We address a problem for which it is known that a set of variables X causes another set of variables Y (e.g., mutations in DNA cause changes in gene expression), and these causal relationships are encoded by a causal network among a set of an unknown number of latent variables. We develop a deep learning model, redundant input neural network (RINN), with a modified architecture and an L1 regularized objective function to find causal relationships between input (X), latent, and output (Y) variables. RINN allows input variables to directly interact with all latent variables in a neural network to influence the information latent variables encode. In a series of simulation experiments, we show that RINN recovers latent causal structure from various simulated datasets better than other models. We hypothesize that training RINN on omics data will enable us to map the functional impacts of genomic alterations to latent variables in a deep learning model, allowing us to discover the hierarchical causal relationships between variables perturbed by different genomic alterations. Importantly, the direct connections between input and latent variables in RINN make the latent variables partially interpretable, as they can be easily mapped to input space. We apply RINN to cancer genomic data, where it is known that genomic alterations cause changes in gene expression. We show that gene expression can be predicted from genomic alterations with reasonable AUROCs. We also show that RINN is able to discover real cancer signaling pathway relationships, especially relationships in the PI3K, Nrf2, and TGFβ pathways, including some causal relationships. However, despite high regularization, the learned causal graphs are still somewhat too dense. We discuss promising future directions for RINN, including differential regularization, autoencoder pretrained representations, and constrained evolutionary strategies.
Share
Citation/Export: |
|
Social Networking: |
|
Details
Item Type: |
University of Pittsburgh ETD
|
Status: |
Unpublished |
Creators/Authors: |
|
ETD Committee: |
|
Date: |
5 June 2020 |
Date Type: |
Publication |
Defense Date: |
20 February 2020 |
Approval Date: |
5 June 2020 |
Submission Date: |
7 March 2020 |
Access Restriction: |
1 year -- Restrict access to University of Pittsburgh for a period of 1 year. |
Number of Pages: |
146 |
Institution: |
University of Pittsburgh |
Schools and Programs: |
School of Computing and Information > Intelligent Systems Program |
Degree: |
PhD - Doctor of Philosophy |
Thesis Type: |
Doctoral Dissertation |
Refereed: |
Yes |
Uncontrolled Keywords: |
Deep Learning
Neural Network
Causality
Cancer
Signaling Pathway
Latent Variable |
Date Deposited: |
05 Jun 2020 21:22 |
Last Modified: |
05 Jun 2021 05:15 |
URI: |
http://d-scholarship.pitt.edu/id/eprint/39022 |
Available Versions of this Item
-
DEEP LEARNING FOR CAUSAL STRUCTURE LEARNING APPLIED TO CANCER PATHWAY DISCOVERY. (deposited 05 Jun 2020 21:22)
[Currently Displayed]
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
![View Item View Item](/style/images/action_view.png) |
View Item |