Publication Date

2018-09-22

Availability

Open access

Embargo Period

2018-09-22

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PHD)

Department

Biochemistry and Molecular Biology (Medicine)

Date of Defense

2018-08-24

First Committee Member

Stephen Lee

Second Committee Member

Feng Gong

Third Committee Member

Shanta Dhar

Fourth Committee Member

Feng- Chun Yang

Fifth Committee Member

Brian Berman

Abstract

This dissertation herein describes a novel strategy for Prostate Cancer (PCa) biomarker discovery employing use of systems biology causal modeling using omics, molecular phenotyping, and Bayesian artificial intelligence. Three studies were conducted towards this goal. In Study 1, a novel strategy that combines biological outputs with Bayesian network learning to identify potential biomarkers for prostate cancer was utilized. This methodology identified two proteins, filamin B (FLNB) and keratin-19, as potential biomarkers for prostate cancer. The network map also identified a direct linkage between FLNB and filamin A (FLNA), a protein previously identified as playing a role in prostate cancer etiology. The proteins identified from the in silico network model were then biologically validated by examining their levels in a panel of prostate cancer cell lines and in human plasma samples. Given the roles FLNA, FLNB, and KRT19 play in prostate cancer and their identification and validation in Study 1, a novel panel of serum biomarkers was developed in Study 2, using a multi-omic approach that defined FLNA, FLNB, and KRT19 in the panel. New ELISAs for FLNA and FLNB, and IPMRM for FLNA were developed and analytically validated by quantitative measurements of the biomarkers in serum. Retrospectively collected and clinically annotated serum samples with PSA values and Gleason scores from subjects who underwent prostate biopsy, and showed no evidence of cancer, with or without indication of prostatic hyperplasia, or had a definitive pathology diagnosis of prostatic adenocarcinoma were analyzed. Probit linear regression models were used to combine the analytes into score functions to address the following clinical questions. Does the biomarker test augment PSA for population screening? Can aggressive disease be differentiated from lower risk disease, and finally can the panel discriminate between prostate cancer and benign prostate hyperplasia? Modeling of the data showed that the new prostate biomarkers and PSA in combination were better than PSA alone in identifying prostate cancer, improved the prediction of high and low risk disease, and improved prediction of cancer versus benign prostate hyperplasia. Building upon Study 2, in Study 3 the utility of a Prostate Cancer Biomarker panel test on the analysis of 777 patients assessed the combinatorial power of filamin-A (FLNA), age, and prostate volume in predicting clinical segregation of BPH versus PCa. Retrospective analysis of biobank samples from patients with LUTS/BPH and patients with PCa was conducted with results indicating a diagnostic performance that is improved over that of PSA alone in discriminating patients with LUTS/BPH from patients with PCa. Use of this panel as a diagnostic test may reduce the number of unnecessary biopsies performed while providing physicians enhanced clinical decision support in addition to more definitive treatment clarity for patients in the future.

Keywords

Bayesian artificial intelligence, Systems biology and Omics; Biomarkers; Prostate Cancer and Benign prostate hypertrophy; keratin-19; filamin-A; filamin-B

Share

COinS