Statistics, Department of

 

Department of Statistics: Dissertations, Theses, and Student Research

First Advisor

Souparno Ghosh

Committee Members

Erin Blankenship, Reka Howard, Lynette Smith

Date of this Version

8-2025

Document Type

Dissertation

Citation

A dissertation presented to the faculty of the Graduate College at the University of Nebraska in partial fulfilment of requirements for the degree of Doctor of Philosophy

Major: Statistics

Under the supervision of Professor Souparno Ghosh

Lincoln, Nebraska, August 2025

Comments

Copyright 2025, Sarah Josephine Aurit. Used by permission

Abstract

An activity cliff (AC) occurs when drugs close in chemical space produce dissimilar biological results. We focus on developing an inferential procedure to detect the presence of ACs in a chemical landscape. If detected, we provide a distance-based procedure that can be used to identify regions of stability in the chemical landscape of interest and generate prediction with higher precision in those areas of stability. We conceptualize the chemical landscape as a spatial random field and use spatial models for prediction of efficacy for new drugs based on “distance” in chemical space. We argue that an AC manifests itself by inducing non-stationarity in the foregoing spatial random field. We utilize a formal non-parametric test of stationarity to detect the presence of ACs. If non-stationarity is detected, a metric-learning algorithm is employed to transform the coordinate system of the original random field. Once completed, the data are retested for stationarity. If the transformed random field is stationary, then ordinary kriging is used for predictions of test points. We show that the precision of the prediction can be further improved by generating convex clusters in the chemical landscape and training cluster-specific spatial models. Finally, we use Euler-Bernoulli beam theory to attach uncertainty to test points that fall outside all clusters.

Advisor: Souparno Ghosh

Download

Share

COinS