Benchmark on Indexing Algorithms for Accelerating Molecular Similarity Search.

Area of Science:

• Computational chemistry
• Cheminformatics
• Algorithm benchmarking

Background:

• Molecular similarity searches are crucial for retrieving analogous compounds from large chemical databases.
• Exhaustive similarity searches are computationally expensive, and existing indexing algorithms often lack Tanimoto similarity metric implementation.
• There is a need for robust benchmarking systems to evaluate computational algorithms for molecular similarity searching.

Purpose of the Study:

• To implement Tanimoto similarity search using recent indexing algorithms (Hnsw, Onng) in Python or C++.
• To develop a benchmark system for evaluating the performance of these indexing algorithms in molecular similarity searching.
• To provide containerized (Docker, Singularity) benchmark systems for accessibility and reproducibility.

Main Methods:

• Implementation of Tanimoto similarity search with Hnsw and Onng indexing algorithms.
• Development of two separate benchmark systems using Docker and Singularity container technologies.
• Evaluation of search effectiveness and efficiency of different indexing algorithms.

Main Results:

• Graph-based methods (Hnsw, Onng) demonstrated superior performance in molecular similarity searching.
• These methods achieved the optimal trade-off between search effectiveness and efficiency.
• The developed benchmark systems are extensible for new algorithms and datasets.

Conclusions:

• Recent indexing algorithms, particularly graph-based ones, significantly enhance molecular similarity search performance.
• The provided benchmark systems facilitate the evaluation and comparison of computational algorithms in cheminformatics.
• The open-source code enables further research and development in efficient chemical database searching.