Prediction of Drug Targets based on In Vitro Activity Profiles Toward Drug Repurposing for Rare Diseases

Area of Science:

• Pharmacology
• Computational Biology
• Toxicology

Background:

• Over 300 million people worldwide have rare diseases, often with limited treatment options.
• Drug repurposing offers a viable strategy for discovering new treatments by identifying new uses for existing drugs.
• Predicting gene targets for chemical compounds is crucial for advancing drug repurposing efforts.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting gene targets of chemical compounds.
• To extend previous work on identifying enriched genes for compounds in the Toxicology in the 21st Century (Tox21) library.
• To facilitate drug repurposing for rare diseases by uncovering novel gene-compound relationships.

Main Methods:

• Machine learning models including Support Vector Machine, K-Nearest Neighbors, Random Forest, and extreme gradient boosting (XGBoost) were developed.
• Tox21 bioassay screening data was utilized for training and evaluating the predictive models.
• Four multi-label prediction embedding algorithms were tested with the XGBoost model: Binary Relevance, Label Powerset, Classifier Chain, and Multi-Output Classifier.

Main Results:

• All four developed machine learning models demonstrated strong performance with an f1-score exceeding 0.7.
• The extreme gradient boosting (XGBoost) model achieved the best performance among the evaluated models.
• The study successfully explored a reliable method for predicting potential gene targets from in vitro activity profiles.

Conclusions:

• Machine learning models, particularly XGBoost, can reliably predict gene targets from in vitro activity data.
• This approach supports drug repurposing by identifying potential new uses for existing compounds.
• The findings contribute to the discovery of novel treatments for rare diseases.