Benchmarking heterogeneous network-based methods for drug repurposing.

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Drug repurposing (DR) offers a cost-effective approach to discover new drug applications.
• Heterogeneous network-based methods show promise for DR by leveraging complex biological interactions.
• A lack of comprehensive benchmarking hinders the assessment of reliability and generalizability of these DR methods.

Purpose of the Study:

• To systematically evaluate and compare ten advanced heterogeneous network-based drug repurposing methods.
• To assess method performance across diverse datasets, including newly introduced ones.
• To analyze the impact of data sparsity and cross-validation strategies on performance metrics.

Main Methods:

• Evaluation of ten drug repurposing methods: matrix factorization (NMF, NMF-PDR, NMF-DR, VDA-GKSBMF), matrix completion (BNNR, OMC, HGIMC), recommendation systems (IBCF, LIBMF), and deep learning (DRDM).
• Systematic benchmarking across eight datasets (six public, two new drug-disease datasets).
• Performance assessment using Area Under the ROC Curve (AUC) and Area Under the Precision-Recall Curve (AUPR).

Main Results:

• OMC consistently demonstrated the highest AUC and AUPR across most datasets.
• BNNR, DRDM, HGIMC, VDA-GKSBMF, and NMF-PDR showed competitive performance.
• NMF-PDR outperformed other NMF-based methods; cross-validation strategies significantly impact AUPR, with prior studies potentially overestimating performance.

Conclusions:

• This work establishes a reliable benchmarking framework and provides new datasets for drug repurposing research.
• OMC emerges as a top-performing method for heterogeneous network-based drug repurposing.
• Standardized evaluation protocols, particularly concerning negative instance handling in cross-validation, are crucial for accurate performance assessment in DR.