Researchers develop 'explainable AI framework' to identify disease pathways and aid drug development

Yoon Sung-il (left), a professor at Chung-Ang University, and Yang Si-young, a professor at Sungkyunkwan University, who led the research. [National Research Foundation of Korea]
Yoon Sung-il (left), a professor at Chung-Ang University, and Yang Si-young, a professor at Sungkyunkwan University, who led the research. [National Research Foundation of Korea]

A new AI system capable of precisely predicting disease signaling pathways and identifying the key genes behind them has been developed by South Korean researchers.

The National Research Foundation of Korea announced that a joint research team — led by Yoon Sung-il of Chung-Ang University, Yang Si-young of Sungkyunkwan University and Cho Chan-mi of Hanyang University — has developed an AI analysis tool called SaintGSE. The system uses large-scale transcriptomic data to precisely predict disease-associated pathways and identify the key genes driving each pathway, along with the evidence supporting those predictions.

Advances in next-generation sequencing technology have produced vast amounts of transcriptomic, or gene expression, data, yet connecting that data to the molecular mechanisms and signaling pathway changes underlying disease has remained a persistent challenge. Complex conditions such as osteoarthritis are particularly difficult to analyze, as multiple biological mechanisms — including aging, inflammation and cartilage degradation — operate simultaneously, requiring tools that can both pinpoint relevant signaling pathways and identify the genes responsible.

There has been a pressing need for explainable AI analysis technology that can autonomously identify disease-related pathways within massive datasets and quantitatively demonstrate, at the gene level, the basis for its conclusions.

The team built SaintGSE by combining an autoencoder with a transformer architecture, enabling the model to precisely predict whether disease-related signaling pathways are activated from complex transcriptomic data patterns. The researchers also incorporated explainable AI, or XAI, techniques so the system can quantify and present each gene's contribution when the AI predicts a disease pathway.

When the team applied the model to transcriptomic data from osteoarthritis and natural compound treatments, it not only accurately interpreted the molecular mechanisms underlying disease onset but also demonstrated the ability to map how treatment candidates act within the body.

SaintGSE prediction results and verification of natural compounds' potential to alleviate osteoarthritis through animal testing. [Chung-Ang University]
SaintGSE prediction results and verification of natural compounds' potential to alleviate osteoarthritis through animal testing. [Chung-Ang University]

While conventional analysis methods have focused primarily on assessing the statistical enrichment of gene lists, SaintGSE predicts pathway-level changes based on whole-transcriptome patterns and can quantitatively identify the genes contributing to each prediction. The researchers said this opens up potential applications beyond sample-specific disease mechanism interpretation — including biomarker candidate discovery, therapeutic target identification and analysis of the mechanisms by which drugs or natural compounds act.

"This research used AI to extract, in an explainable way, the key pathways and causative genes driving disease from complex genetic data," Yoon said. "It can be applied to interpreting the molecular mechanisms of complex diseases, discovering biomarkers, identifying new therapeutic targets, analyzing drug responses, and predicting the mechanisms of natural compounds or drug candidates."

The study, conducted under a basic research laboratory support project funded by the Ministry of Science and ICT and the National Research Foundation of Korea, was published online April 16 in the international journal Osteoarthritis and Cartilage.


nbgkoo@heraldcorp.com