Causality Discovery Methods for Studying Drug-Drug Interactions and Adverse Effects

The School of EECS is hosting the following PhD Progress Review 3 Thesis Review Seminar:

Causality Discovery Methods for Studying Drug-Drug Interactions and Adverse Effects

Speaker: Sitthichoke Subpaiboonkit
Host: Prof. Guido Zuccon 

Abstract: Causality discovery spans various disciplines, including medicine, pharmacology, biology, economics, and social science. It involves identifying the connections between cause-and-effect events, leading to a deeper understanding of the mechanisms behind these events. This task is critical for advancing knowledge and applications in these fields. Traditionally, causality is determined through a gold standard method known as a randomized controlled trial. However, this approach is often impractical due to its high costs, ethical issues, and time constraints.

Causal Bayesian Network is a leading computational method for uncovering causality in observational data. Alternatively, the constraint-based causal discovery approach examines only a sub-graph of the entire Bayesian Network. However, both methods are computationally intensive. This thesis aims to discover causality in drug-drug interaction (DDI) identification, which is essential for drug safety surveillance. Identifying DDIs is challenging due to the high time complexity of existing causal methods and the absence of a ground truth dataset.

In the first part of this thesis, we proposed Domain-Knowledge-Driven Causality Discovery (DCD) for identifying DDI. Our approach employs external knowledge bases of drug interactions and adverse effects to guide the identification of other unknown interacting drugs. To address this challenge, we employed a constraint-based greedy algorithm combined with a pruning process based on conditional independence.

In the second part, we introduced the causality discovery method to identify multiple causes and combined causes in DDI. We proposed a new concept based on Bayesian V-structure as a rule for discovering these DDI properties. A heuristic algorithm was devised and applied to reduce time complexity by limiting the scope to two interacting drugs.

To accelerate drug-drug interaction research, we finally introduce a newly created dataset with ground truth. In this work, we use MetaMap to map the original dataset, which lacks ground truth, with knowledge from drug-drug interaction databases, thereby deriving the ground truth dataset.

Biography: Sitthichoke Subpaiboonkit received his B.S. degree in Computer Science and M.S. degree in Bioinformatics from Chiang Mai University, Thailand. Currently, he is pursuing a PhD at the School of Electrical Engineering and Computer Science, University of Queensland, under the supervision of Prof. Guido Zuccon as the main supervisor and Prof. Xue Li as the co-supervisor. His research interests include machine learning, causality discovery, and bioinformatics.