Explainable artificial intelligence in high-throughput drug repositioning for subgroup stratifications with...

Area of Science:

• Computational biology
• Bioinformatics
• Artificial Intelligence in Medicine

Background:

• Precision medicine necessitates patient stratification and targeted therapies for heterogeneous diseases.
• De novo drug development is costly, time-consuming, and has low FDA approval rates, making it unfeasible for small patient subgroups.
• Drug repositioning offers a viable alternative for identifying new treatments for specific patient subpopulations.

Purpose of the Study:

• To develop an explainable AI approach for patient stratification and drug repositioning.
• To identify druggable homogeneous subgroups within a heterogeneous disease population.
• To reposition existing drugs for identified patient subgroups.

Main Methods:

• Utilized contrast pattern mining and network analysis to discover homogeneous patient subgroups.
• Performed biomedical network analysis for each subgroup to identify relevant drugs.
• Ranked candidate drugs using an aggregated drug score within a human-in-the-loop framework.
• Employed phenotypic and genotypic patient data with a heterogeneous knowledge base for colorectal cancer (CRC).

Main Results:

• Identified homogeneous patient subgroups within the CRC population.
• Discovered a set of relevant and ranked candidate drugs for each subgroup.
• Validated that top candidate drugs are cancer-related and potentially applicable as CRC regimens.
• Demonstrated the utility of a multi-view perspective using phenotypic and genotypic data.

Conclusions:

• The developed explainable AI approach effectively stratifies patients and identifies drug repositioning candidates.
• The human-in-the-loop framework facilitates expert hypothesis generation and insight discovery.
• This data-driven strategy supports precision medicine by finding tailored therapies for specific patient subgroups.
• The study highlights the potential of drug repositioning for addressing unmet needs in diseases like colorectal cancer.