PhenoBIC: operator-free single-cell spatial phenotyping in multiplex imaging data using deep learning of cell...

Area of Science:

• Computational biology
• Biomedical imaging
• Machine learning

Background:

• Multiplex imaging enables single-cell spatial analysis of tissue microenvironments.
• Current cell phenotyping in multiplex imaging is manual, time-consuming, and operator-dependent.

Purpose of the Study:

• To develop an automated, accurate, and robust computational method for cell phenotyping in multiplex imaging.
• To introduce PhenoBIC (Biomarker Imprint of a Cell), a deep learning model for classifying cell phenotypes based on multiplexed biomarker signals.

Main Methods:

• Development of PhenoBIC, a pre-trained deep learning model for image classification.
• Training and validation using a dataset of ~1.4 million manually curated cell expression labels.
• Comparative analysis against manual gating and other machine learning approaches.

Main Results:

• PhenoBIC achieved a high F1-score (~0.88) for cell marker expression classification.
• The model demonstrated robust performance across diverse biomarkers, tissue types, and imaging platforms.
• PhenoBIC outperformed manual gating and existing computational methods.

Conclusions:

• PhenoBIC offers a significant advancement in automated cell phenotyping for multiplex imaging analysis.
• The open-sourced model and datasets facilitate wider adoption and reproducibility in biological and clinical research.
• Automated phenotyping accelerates the discovery of insights from complex tissue microenvironments.