Date of Award

3-24-2022

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Bioinformatics

First Advisor

Minjun Chen

Second Advisor

Mary Yang

Abstract

Drug-induced liver injury (DILI) is a rare adverse reaction that can result in liver transplant or death. Currently, there are no reliable methods to identify which individuals are at risk of developing severe or chronic DILI prior to drug administration. Here, we investigated the interplay between drug properties, clinical, and genetic factors and DILI risk, severity, and trajectory. First, we used a large, published dataset of 1036 drugs to examine the relationship between DILI risk and two drug properties, lipophilicity and extent of metabolism. Each drug property was tested alone and in combination with daily dose using five different DILI risk annotations. Both lipophilicity and extent of metabolism were significantly associated with DILI risk in some of the DILI annotations. When combined with daily dose, both were significant across all annotations. Therefore, these drug properties are important risk factors for DILI, and the association is stronger when dose is included. Next, we investigated factors influencing the trajectory of DILI recovery. Using accelerated failure time to analyze 294 cases collected by the International Drug-Induced Liver Network Consortium (IDILIC), we found that a longer time to onset, non-significant drug metabolism, and higher serum bilirubin and alkaline phosphatase level at DILI onset were associated with a longer time to recovery. These factors were included in a multivariate model which identified high-and low-risk groups. For the high-risk group, the estimated probability of recovery by 6 months was 0.46 (95% CI 0.26–0.61) versus 0.93 (95% CI 0.58–0.99) for the low-risk group. The model performance was validated in two independent cohorts. The recovery trajectory differed significantly between high- and low-risk groups, with the majority of low-risk cases recovering sooner. Finally, we constructed machine learning models to evaluate the relationship between DILI severity and drug properties, clinical factors, and imputed HLA type. The best performing model achieved a mean balanced accuracy of 0.67 during cross-validation. Our findings offer important insights into the factors influencing DILI risk, severity, and recovery. Identifying characteristics that define which patient subpopulations at a higher risk of can help minimize future risk of adverse reactions.

Download

Share

COinS