Supervised-Learning-Driven Interrogation of Organ-on-a-Chip Quality from Microscopy Images

Area of Science:

• Biotechnology
• Machine Learning
• Cell Biology

Background:

• Organ-on-a-chip models are crucial for drug discovery and disease modeling.
• Objective quality control of these complex microphysiological systems is essential but challenging.
• Microscopy images are a primary source of data for assessing model quality.

Purpose of the Study:

• To develop a high-throughput, objective method for interrogating organ-on-a-chip quality using supervised learning.
• To assess the performance of machine learning classifiers in distinguishing cell types and quality in lung-on-a-chip models.
• To investigate the impact of dimensionality reduction on classifier performance and computational efficiency.

Main Methods:

• Collected over 600 microscopy images of two lung cell types in lung-on-a-chip models.
• Trained supervised learning classifiers to predict cell types and model quality.
• Applied dimensionality reduction techniques to enhance classifier performance and reduce computational load.

Main Results:

• Trained classifiers achieved >95% AUC for cell type prediction and >83% accuracy for quality assessment.
• Dimensionality reduction improved predictive capacity for some classifiers.
• Computational costs for resource-intensive algorithms were significantly reduced.

Conclusions:

• Supervised learning provides a robust and objective method for automated organ-on-a-chip quality control.
• Machine learning integration, including dimensionality reduction, can optimize the development and implementation of organ-on-a-chip technologies.
• This approach is expected to accelerate the adoption of machine learning in automating organ-on-a-chip development.