Machine Learning-Assisted High-Content Imaging Analysis of 3D MCF7 Microtissues for Estrogenic Effect Prediction

Area of Science:

• Toxicology and Environmental Health
• Cell Biology and Imaging
• Computational Biology and Bioinformatics

Background:

• Endocrine-disrupting chemicals (EDCs) present significant environmental and health risks, but high-throughput screening is challenging.
• Three-dimensional (3D) cell cultures offer more physiologically relevant models for studying chemical effects.
• Quantitative imaging techniques can capture complex cellular responses in 3D systems.

Approach:

• Established a 3D imaging and analysis pipeline for MCF-7 breast cancer cell microtissues exposed to estradiol (E2) and propyl pyrazole triol (PPT).
• Developed a machine learning model using differential image features to classify EDC exposure.
• Created deep learning software for automated characterization of microtissue lumen formation.

Key Points:

• The machine learning model achieved high accuracy (AUC-ROC ~0.95) in distinguishing between E2 and PPT exposure based on 140 image features.
• Automated software accurately quantified lumen formation, a key indicator of estrogenic activity.
• Integrated image feature profiling and lumen characterization captured functional estrogen receptor signaling complexity.

Conclusions:

• An integrated 3D imaging, machine learning, and deep learning approach provides a robust method for screening estrogenic EDCs.
• This framework offers a promising conceptual strategy for assessing the risks posed by endocrine-disrupting chemicals.
• The study demonstrates the utility of advanced imaging and computational methods in toxicological risk assessment.