Estimating error rates in bioactivity databases

Area of Science:

• Drug Discovery
• cheminformatics
• Bioactivity Data Analysis

Background:

• Bioactivity databases are essential for drug discovery, aiding target prediction for small molecules.
• Manual curation from scientific literature and patents introduces human errors, potentially leading to flawed early-stage drug discovery decisions.

Purpose of the Study:

• To compare bioactivity data from ChEMBL, Liceptor, and WOMBAT databases using identical source documents.
• To estimate error rates for various data parameters and identify databases with higher error frequencies.
• To highlight the impact of data quality on drug discovery and guide data curation efforts.

Main Methods:

• Comparative analysis of curated bioactivity data from three major databases (ChEMBL, Liceptor, WOMBAT).
• Estimation of error rates for specific parameters including small molecule structures, target information, activity values, and activity types.
• Analysis of supplier-specific error rates to identify potential curation inconsistencies.

Main Results:

• Small molecule structures exhibit the highest estimated error rate, followed by target, activity value, and activity type.
• Error rate patterns are consistent across supplier-specific estimates.
• Identified specific data points and parameter types that are most susceptible to errors, aiding in targeted data re-curation.

Conclusions:

• The study provides crucial error rate estimates for key bioactivity data parameters across major databases.
• Awareness of error frequencies and types in bioactivity data is vital for scientists to mitigate risks in drug discovery.
• The findings support the need for improved data curation standards and targeted re-curation efforts to enhance data reliability.