Lee, Sanghoon
(2021)
Integral Genomic Signature Modeling in Breast Cancer to Predict HER2-targeted Therapy Response.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
While many effective therapies have been developed for breast cancer, overtreatment of clinically localized or regional tumors remain a major clinical problem. In particular, HER2-positive breast cancer patients are treated with the combination of HER2-targeted therapy and chemotherapy, but the clinical responses are discordant. With the advent of low-cost genome sequencing, breast oncology is expected to undergo a deep transformation by leveraging multi-omics sequencing to guide precision treatment decisions, which is deemed to be highly cost-effective. The relationship between genomic features and the therapeutic responses lays the foundation to maximize the effect of the therapeutic treatment based on a patient’s genomic context. However, current big-data based modeling methods are plagued with insufficient cross-dataset performance against experimental variations and sequencing bias, and lack of biological relevance. Identification of predictive genomic signatures for drug sensitivity and developing a precise predictive algorithm hold the key to optimizing the decision for effective intervention and the prediction of clinical outcomes in the era of precision oncology.
The goals of our research are to build a therapeutic response prediction model using multi-omics data extracted from breast cancer patients and to validate the model in multiple clinical trial datasets. We postulate that the redundancy within high-dimensional genomic features, which are typically eliminated via dimensionality reduction or feature removal during multi-omics modeling, may help strengthen the predictive powers during cross-dataset modeling. This concept is similar to the use of redundant steel rods to reinforce the pillars of a building. Based on this principle, we propose an integral genomic signature (iGenSig) analysis that leverages high-dimensional redundant genomic features to strengthen the predictive “pillar”, which we termed as an integral genomic signature, and adaptively resolve feature redundancies within the pillar. The iGenSig-HT model can be applied to predict patient response in independent validation datasets with outstanding cross-dataset applicability and resilience against simulated errors in genomic features compared to machine learning and deep learning methods.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
|
| ETD Committee: |
|
| Date: |
21 August 2021 |
| Date Type: |
Publication |
| Defense Date: |
29 April 2021 |
| Approval Date: |
21 August 2021 |
| Submission Date: |
27 May 2021 |
| Access Restriction: |
2 year -- Restrict access to University of Pittsburgh for a period of 2 years. |
| Number of Pages: |
112 |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
School of Medicine > Biomedical Informatics |
| Degree: |
PhD - Doctor of Philosophy |
| Thesis Type: |
Doctoral Dissertation |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
HER2-targeted therapeutic response prediction, HER2-positive breast cancer patients, BCL2L14-ETV6 gene fusion |
| Date Deposited: |
22 Aug 2021 02:35 |
| Last Modified: |
21 Aug 2023 05:15 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/41188 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}