Single-cell hit calling in high-content imaging screens with Buscar.

Area of Science:

• Computational Biology
• Cellular Imaging
• Bioinformatics

Background:

• High-content screening (HCS) quantifies single-cell morphology but often uses aggregate statistics, masking subpopulation effects.
• Current image-based profiling methods assume cell homogeneity, reducing sensitivity to subtle or heterogeneous perturbation effects.

Purpose of the Study:

• Introduce Buscar, a novel method for hit calling in HCS that utilizes single-cell heterogeneity.
• Enable simultaneous quantification of perturbation efficacy and specificity by analyzing morphology states.

Main Methods:

• Buscar compares two reference single-cell populations (e.g., disease vs. healthy) to define morphology signatures.
• These signatures are used to score perturbations, quantifying efficacy and off-target effects.
• The method leverages the full heterogeneity of image-based profiles, avoiding aggregation bias.

Main Results:

• Buscar successfully quantified morphology rescue and off-target activity in a cardiac fibroblast dataset.
• The method recovered biologically relevant gene-phenotype associations in the MitoCheck dataset.
• Buscar demonstrated robustness across technical replicates in small-molecule and CRISPR-Cas9 perturbations (CPJUMP1 dataset).

Conclusions:

• Buscar provides a reproducible and interpretable hit-calling method for HCS.
• It overcomes aggregation bias, enhancing sensitivity and enabling simultaneous assessment of efficacy and specificity.
• Buscar is released as an open-source Python package to facilitate its adoption.