DDA-bench: a manually curated database for benchmarking datasets and baseline performance values in predicting...

Area of Science:

• Computational biology
• Pharmacogenomics
• Bioinformatics

Background:

• Predicting drug-disease associations aids in novel drug development.
• Existing computational methods require reliable benchmarking datasets and baseline performance data.
• Manual curation of these resources is labor-intensive and time-consuming.

Purpose of the Study:

• To develop DDA-Bench, a database service for drug-disease association prediction.
• To curate commonly used benchmarking datasets and up-to-date performance values from baseline studies.
• To analyze data records and identify factors influencing predictive performance.

Main Methods:

• Development of the DDA-Bench database service.
• Curation of benchmarking datasets for drug-disease association studies.
• Extraction and compilation of performance values from published baseline studies.
• Analysis of database records to assess performance variations and dataset characteristics.

Main Results:

• The DDA-Bench database consolidates essential resources for predictive modeling.
• Analysis revealed significant performance variations for methods across different reports.
• Dataset density was identified as a critical factor impacting predictive performance.

Conclusions:

• DDA-Bench significantly reduces the time and effort required for data preparation in drug-disease association research.
• Attention to performance variations and dataset density is essential for robust model development.
• The DDA-Bench database is publicly accessible to facilitate research in the field.