Integrating artificial intelligence and manual curation to enhance bioassay annotations in ChEMBL

Area of Science:

• Cheminformatics
• Bioactivity data analysis
• Machine learning applications

Background:

• The ChEMBL database contains a growing volume and diversity of bioactivity data.
• Standardized, interoperable, and machine-readable assay metadata is critical for cheminformatics and machine learning (ML) applications.
• Current ChEMBL bioassay annotations require enhancement for improved data quality and granularity.

Purpose of the Study:

• To present recent efforts to enhance the quality and granularity of bioassay annotations in the ChEMBL database.
• To improve the standardization, interoperability, and machine readability of assay metadata.
• To enable more robust downstream analyses and precise compound-target activity modeling.

Main Methods:

• Implemented a "perfect assay description" template for consistent annotation.
• Utilized natural language processing (NLP) techniques and multi-class classification for automated extraction of assay parameters and categorization of legacy data.
• Developed and validated a spaCy-based Named Entity Recognition (NER) model to identify experimental methods.
• Created a complementary classification model to refine ASSAY_TYPE categorization.
• Improved metadata extraction for ADME endpoints, organism and protein variant annotations, and ontology linking using tools like text2term.

Main Results:

• The "perfect assay description" template guides consistent annotation practices.
• NLP and classification models automatically extract key assay parameters and assign broad assay categories.
• The spaCy-based NER model achieves high precision and recall in identifying experimental methods.
• A complementary classification model enhances ASSAY_TYPE categorization beyond the existing schema.
• Improvements in metadata extraction and ontology linking advance the FAIRness of ChEMBL bioassay data.

Conclusions:

• Recent enhancements significantly improve the quality and granularity of ChEMBL bioassay annotations.
• These improvements advance the Findable, Accessible, Interoperable, and Reusable (FAIR) principles for ChEMBL bioassay data.
• The enhanced data enables more robust downstream analyses and more precise compound-target activity modeling in cheminformatics and ML applications.