Researchers develop machine learning model to identify disease subtypes

Researchers at the Hebrew University of Jerusalem have developed a machine learning approach to identify potential subtypes of diseases, significantly enhancing the field of disease classification and treatment strategies. Led by doctoral student Dan Ofer and Professor Michal Linyar of the Department of Biochemistry at the Hebrew University's Institute of Life Sciences, the study marks a major advance in the use of artificial intelligence in medical research.

The survey results are as follows: Biomedical Informatics Journal.

Differentiating diseases into distinct subtypes is crucial for precise research and effective treatment strategies. The Open Targets Platform integrates biomedical, genetic, and biochemical datasets to support disease ontology, classification, and potential gene targets. However, annotation of many diseases remains incomplete and often requires extensive expert medical input. This challenge is particularly critical for rare and orphan diseases where resources are limited.

In this study, we introduce a novel machine learning approach to identify diseases with potential subtypes. We leveraged an extensive database of approximately 23,000 diseases recorded in the Open Targets Platform to derive novel features for predicting diseases with subtypes using direct evidence. We then applied machine learning models to analyze the importance of features and evaluate the predictive performance, revealing both known and novel disease subtypes.

The model achieved an impressive 89.4% area under the receiver operating characteristic curve in identifying known disease subtypes. The integration of a pre-trained deep learning language model further improved the model's performance. Notably, the study identified 515 candidate diseases predicted to have previously unannotated subtypes, paving the way for new insights into disease classification.

“This project demonstrates the incredible potential of machine learning to improve our understanding of complex diseases,” Offer said.

“By utilizing advanced models, we can uncover previously hidden patterns and subtypes, ultimately contributing to more precise and personalized treatment.”

This innovative methodology enables a robust and scalable approach to improve knowledge-based annotation, allowing for a comprehensive evaluation of disease ontology layers.

“We are excited by the potential of our machine learning approach to revolutionize disease classification,” said Professor Linial. “Our findings could make a significant contribution to personalized medicine and open new avenues for therapeutic development.”