AI-driven phenotypic profiling of organoids

Organoids have revolutionized in vitro modeling by accurately replicating native tissue structure and producing drug response results that better reflect real-life scenarios. However, effectively visualizing and quantifying organoid characteristics still poses challenges. The researchers introduce Morphi, an artificial intelligence framework that infers organoid health and state directly from bright-field microscopy images, using machine learning technology.

Morphi utilizes fluorescence-based viability signals as weak supervision to train a pixel-level viability model, eliminating the need for dense, manual annotation. The model accurately segments each organoid from an image and produces a percentage viability score along with a spatial death map, revealing dead sub-organoid regions.

This approach captures graded toxic responses at single-organoid resolution and works effectively across various tissue types, offering a scalable, label-free method for organoid phenotypic profiling in large-scale drug screening. Morphi further integrates these viability predictions with unsupervised features derived from morphology and texture analysis, creating a high-dimensional phenotypic space for each organoid.

The resulting phenotypic fingerprints capture individual organoid states and their population-level distributions across different perturbation conditions. These fingerprints effectively cluster drugs based on their mechanism of action, distinguish pharmacologically related drugs, and uncover phenotypic convergence across diverse perturbagens.

Ultimately, Morphi provides a comprehensive, quantitative readout of organoid viability and phenotypic response, facilitating high-content, mechanism-aware drug screening using organoids.