Target enhanced 2D similarity search by using explicit biological activity annotations and profiles.

Area of Science:

• Cheminformatics
• Computational Chemistry
• Drug Discovery

Background:

• Biological activity data in chemical databases like PubChem is a valuable resource.
• Conventional similarity search (CSS) using 2D fingerprints lacks consideration for feature importance.

Purpose of the Study:

• To compare the performance of different 2D similarity search methods.
• To evaluate the impact of profile-based approaches on search efficiency and accuracy.

Main Methods:

• Large-scale investigation of 208 compound activity classes from PubChem.
• Comparison of Conventional Similarity Search (CSS), Iterative Similarity Search (ISS), and Iterative Similarity Search with Classification (ISC).
• Introduction of profile-based enhancements to these search algorithms.

Main Results:

• ISC significantly improves precision compared to CSS and ISS.
• Profile-based searches enhance recall and reduce search time for ISS and ISC without performance loss.
• Profile-based ISS (PBISS) and profile-based ISC (PBISC) are more efficient.

Conclusions:

• ISC and ISS demonstrate superior performance over CSS for similarity searching.
• Profile-based iterative similarity searches are more efficient and perform better than standard algorithms.
• Profile-based algorithms offer a significant advantage in speed and accuracy for large-scale compound screening.