Implicit-descriptor ligand-based virtual screening by means of collaborative filtering.

Area of Science:

• Computational Chemistry
• Cheminformatics
• Machine Learning

Background:

• Ligand-based virtual screening commonly uses molecular fingerprinting for feature extraction.
• Existing methods rely on explicit vectorization, which can introduce bias and struggle with data sparsity.

Purpose of the Study:

• To develop and evaluate novel machine learning methods for virtual screening that avoid explicit molecular fingerprinting.
• To introduce implicit descriptors derived from assay data, inspired by recommendation system algorithms.

Main Methods:

• Utilized collaborative filtering algorithms from recommendation systems to generate implicit ligand descriptors.
• Evaluated performance against traditional machine learning methods that rely on molecular fingerprints.

Main Results:

• Implicit descriptor methods significantly outperformed traditional fingerprint-based machine learning approaches.
• Demonstrated superior resilience to target-ligand data sparsity.
• Showcased a high potential for identifying promiscuous ligands.

Conclusions:

• Implicit descriptors offer a powerful, bias-free alternative to molecular fingerprinting in virtual screening.
• These methods enhance the robustness and predictive power of ligand-based machine learning, particularly in challenging datasets.