Benchmarking feature selection methods for compressing image information in high-content screening.

Area of Science:

• Biopharmaceutical drug discovery
• High Content Screening (HCS)
• Automated microscopy imaging assays

Background:

• High Content Screening (HCS) generates high-dimensional data from automated microscopy.
• Extracting maximal information from HCS requires extensive feature extraction and complex analysis.
• Automated feature selection is crucial for reducing data complexity and enhancing interpretability in drug development.

Purpose of the Study:

• To compare the performance of 12 state-of-the-art feature selection methods (supervised and unsupervised).
• To evaluate the generalizability and importance of selected features using automated machine learning (AutoML).
• To provide practical guidance for experimental design and method selection in HCS-based drug discovery.

Main Methods:

• Systematic testing of 12 feature selection algorithms on two relevant HCS datasets.
• Utilizing automated machine learning (AutoML) for unbiased evaluation of selected features.
• Assessing feature selection performance using standard plate, assay, and compound statistics.

Main Results:

• Identified effective feature selection strategies for HCS data analysis.
• Demonstrated the utility of both supervised and unsupervised methods based on data characteristics.
• Provided insights into optimal feature set sizing for meaningful interpretation.

Conclusions:

• Feature selection significantly impacts the interpretability and utility of HCS data in drug discovery.
• The choice between supervised and unsupervised methods depends on the availability of experimental controls.
• Results offer practical recommendations for optimizing HCS experimental design and data analysis pipelines.