PegaPlus─Interactive Machine Learning by Human Observation for Efficient Clustering and Analysis of...

Area of Science:

• Drug Discovery and Development
• Computational Chemistry
• Bioinformatics

Background:

• Early-stage drug discovery requires identifying lead-like molecular series for structure-activity relationship (SAR) studies.
• Clustering high-throughput screening (HTS) data is a common initial step, followed by expert refinement to identify bioactive compound classes.

Purpose of the Study:

• To present PegaPlus, a novel, interactive, and visual approach for HTS data analysis.
• To facilitate the integration of expert knowledge into HTS data analysis through a learning-by-observation strategy.

Main Methods:

• PegaPlus employs a stochastic proximity embedding algorithm for 2D data visualization via a web interface.
• Users interactively refine compound clustering by manipulating data points, with an online support vector machine learning from these modifications.
• The visualization updates dynamically based on user input and molecular similarity.

Main Results:

• PegaPlus significantly reduces the number of refinement steps by half compared to purely manual methods.
• The interactive machine learning approach effectively supports medicinal chemists in their tasks.
• Automates time-consuming manual refinement processes in HTS data analysis.

Conclusions:

• PegaPlus demonstrates the efficacy of interactive machine learning in streamlining HTS data analysis.
• The tool reliably supports medicinal chemists by automating manual refinement tasks.
• PegaPlus is openly available as a web server for broader accessibility.