Extracting Compound Profiling Matrices from Screening Data.

Area of Science:

• Chemoinformatics and Chemogenomics
• Computational Biology
• Drug Discovery

Background:

• Compound profiling matrices are essential for structure-activity relationship (SAR) studies and chemoinformatic applications.
• Publicly available experimental compound profiling matrices are scarce, hindering research and machine learning model development.
• Existing matrices are often proprietary, limiting broader scientific access and utilization.

Purpose of the Study:

• To develop and present an algorithm for generating large-scale compound profiling matrices.
• To address the scarcity of publicly available experimental profiling data.
• To facilitate machine learning-based multitask activity prediction and chemoinformatic analyses.

Main Methods:

• A novel algorithm, adapted from biclustering techniques used in genomics.
• Iterative removal of entities limiting assay or compound coverage to increase matrix completeness.
• Application to assays from the PubChem BioAssay database.
• Weight settings to control matrix size and prioritize assays or compounds.

Main Results:

• Generation of large profiling matrices (e.g., >100,000 compounds x 50-100 targets).
• Method enables creation of matrices with increasing coverage and enrichment of active entries.
• Successful application to publicly available assay data.

Conclusions:

• The developed algorithm effectively generates large, valuable compound profiling matrices.
• This methodology enhances data availability for chemoinformatics, chemogenomics, and machine learning.
• The approach provides a scalable solution for creating rich datasets for drug discovery research.