Machine learning approach for discrimination of genotypes based on bright-field cellular images

Area of Science:

• Cell biology
• Machine learning
• High-throughput screening

Background:

• Morphological profiling combines microscopy and machine vision for high-throughput phenotyping.
• While fluorescent microscopy is common for single-cell profiling, the utility of bright-field (BF) microscopy for label-free cells is less explored.

Purpose of the Study:

• To investigate if BF microscopy images of label-free cells can discriminate single-gene perturbations.
• To assess the potential of BF microscopy for mutant cell profiling.

Main Methods:

• Acquisition of hundreds of BF images from single-gene mutant and wild-type cells.
• Quantification of single-cell profiles using texture features from cellular regions.
• Development of a machine learning model to differentiate mutant from wild-type cells.

Main Results:

• The machine learning model successfully discriminated mutant cells from wild-type cells with an area under the receiver operating characteristic curve of 0.773.
• Identified key discriminating features related to the morphology of structures within cellular regions.
• Observed similar feature profiles for functionally related gene pairs in mutant cells.

Conclusions:

• BF microscopy images of label-free cells contain sufficient information to discriminate single-gene perturbations.
• Morphological profiling using BF microscopy shows potential as a valuable tool for genetic screening and cell profiling.