Adaptive multi-view learning method for enhanced drug repurposing using chemical-induced transcriptional profiles,...

Area of Science:

• Computational biology and bioinformatics
• Drug discovery and development
• Translational medicine

Background:

• Traditional drug development is costly and time-consuming.
• Existing drug repurposing methods often use limited data and simplistic hypotheses.
• There is a need for advanced computational approaches to integrate complex biological data for drug repurposing.

Purpose of the Study:

• To introduce adaptive multi-view learning (AMVL), a novel methodology for enhanced drug repurposing.
• To integrate chemical-induced transcriptional profiles (CTPs), knowledge graph (KG) embeddings, and large language model (LLM) representations.
• To improve the accuracy and efficiency of predicting drug-disease associations.

Main Methods:

• Adaptive multi-view learning (AMVL) framework.
• Integration of CTPs, KG embeddings, and LLM representations.
• Similarity matrix expansion, multi-view learning (MVL), matrix factorization, and ensemble optimization.
• Evaluation on benchmark (Fdataset, Cdataset, Ydataset) and iDrug datasets.

Main Results:

• AMVL significantly outperforms state-of-the-art (SOTA) methods in predicting drug-disease associations.
• Achieved superior accuracy across multiple metrics on benchmark and large-scale datasets.
• Literature-based validation confirmed predictive capabilities, with 70% of top predictions corroborated by recent evidence.

Conclusions:

• AMVL provides a robust and scalable solution for accelerating drug discovery.
• The methodology effectively integrates diverse data modalities for enhanced drug repurposing.
• Open-sourced data and code promote transparency, reproducibility, and further innovation in translational medicine.