Adapting Document Similarity Measures for Ligand-Based Virtual Screening.

Area of Science:

• Computational Chemistry
• Cheminformatics
• Drug Discovery

Background:

• Quantifying molecular similarity is crucial for virtual screening in drug discovery.
• Existing similarity measures, often adapted from text retrieval, show promise but can be improved.
• Ligand-based virtual screening relies heavily on effective similarity metrics.

Purpose of the Study:

• To introduce a new similarity measure, the Adapted Similarity Measure of Text Processing (ASMTP), for ligand-based virtual screening.
• To evaluate the efficacy of ASMTP against established methods using benchmark datasets.
• To demonstrate the superiority of ASMTP in improving virtual screening performance.

Main Methods:

• Developed the Adapted Similarity Measure of Text Processing (ASMTP) from text processing techniques.
• Tested ASMTP on Maximum Unbiased Validation (MUV) and MDL Drug Data Report (MDDR) datasets.
• Compared ASMTP performance against the Tanimoto coefficient and other similarity measures using recall at 1% and 5% cut-offs.

Main Results:

• The proposed ASMTP measure demonstrated superior performance in ligand-based virtual screening.
• ASMTP outperformed the conventional Tanimoto coefficient in benchmark dataset evaluations.
• Experiments confirmed ASMTP's effectiveness in identifying relevant molecular structures.

Conclusions:

• The Adapted Similarity Measure of Text Processing (ASMTP) is a highly effective novel metric for ligand-based virtual screening.
• ASMTP offers an improved approach compared to existing similarity coefficients, including the Tanimoto coefficient.
• This new measure has the potential to significantly advance virtual screening accuracy and efficiency in drug discovery.