Applying Deep Neural Network Analysis to High-Content Image-Based Assays.

Area of Science:

• Cellular imaging
• Machine learning
• Genetics

Background:

• Central nervous system (CNS) disorders often have unclear causes due to multiple pathological subtypes and complex genetic factors.
• Integrating diverse patient data, including imaging and genetic information, with machine learning offers a path to understanding disease heterogeneity.

Purpose of the Study:

• To investigate if high-content imaging of primary skin fibroblasts using the Cell Painting method can reveal disease-specific information.
• To develop and validate a machine learning model capable of distinguishing between healthy controls and patients with spinal muscular atrophy (SMA).

Main Methods:

• Utilized Cell Painting, a high-content imaging technique, on primary skin fibroblasts.
• Employed a pre-trained deep neural network and deep image embeddings to identify and account for technical nuisance signals in imaging data.
• Developed a convolutional neural network (CNN) using a subset of cells and tested its ability to differentiate unseen control and SMA patient cells, while controlling for batch effects.

Main Results:

• Technical variations (batch/plate type, location) introduced detectable signals but were manageable with appropriate experimental design.
• The developed CNN model successfully differentiated between healthy controls and SMA patients.
• Model performance was robust and not significantly affected by batch or plate type variations.

Conclusions:

• High-content imaging of fibroblasts, combined with machine learning, can effectively stratify patients based on disease state.
• This methodology provides a foundation for studying complex genetic neurological disorders with unknown subtypes.
• Further research is warranted to explore broader applications in CNS disorder subtyping.