Computational approaches to identify circular RNAs in Parkinson's Disease

Authors

• Lisa Shrestha, University of Alabama at Birmingham

Advisor(s)

André Leier

Committee Member(s)

Ashley Harms
David Schneider
Tatiana Marquez-Lago
Tatjana Coric

School

College of Arts and Sciences

Document Type

Dissertation

Department (new version)

Biology

Date of Award

9-9-2024

Abstract

Parkinson’s Disease (PD) is the second most common neurodegenerative disease, affecting approximately a million individuals in the United States. Despite its high prevalence, PD is generally diagnosed at a progressed stage after the manifestation of characteristic motor symptoms (bradykinesia, tremor, rigidity, etc.). Hence, it is imperative to understand the underlying molecular mechanisms involved in the progression of PD at an early stage. Several non-coding RNAs have emerged as important regulators in several key cellular processes, among which circular RNAs' role has not been studied well enough. Here, we explore the changes in the expression of circRNAs with respect to the progression of PD using large-scale longitudinal transcriptomics data and their downstream targets. We also adopt a hybrid deep learning approach to develop a tool that can be utilized for in silico validation of the unannotated circRNAs identified in our analysis. We identified several circRNAs that were found to be differentially expressed in early-stage PD patients in the first few years enrolled in the study, out of which 6 circRNAs (circSLC8A1, circPICALM, circVRK1, circMAN1A2, circZNF91, and circEMB) were further studied and a significant ceRNA network of these circRNAs with their 6 downstream miRNAs with sponging effect (hsa-miR-500a-3p, has-miR-162-3p, hsa-miR-362-5p, hsa-miR-3928-3p, hsa-miR-6815-5p, hsa-miR-1294) and 11 gene targets ((UBL3, OXR1, MRPS10, XPOT, GLO1, TMF1, TMDF3, IPO7, CCT4, BMI1, and ANP32E) associated with PD progression. A multivariate logistic regression model with the change in expression of the circRNAs and gene targets as predictors identified PD patients with clinically pertinent progression with an AUC of 0.70 [0.604 - 0.802]. We also explored using a pre-trained BERT model called DNABERT as a predictor of circRNAs that can be used to computationally confirm unannotated/novel circRNAs identified in PD transcriptomics data as circRNAs. These DL models can classify circRNAs from other long non-coding RNAs.

ProQuest ID

https://proquest.com/docview/3101958595