AI-driven Classification of Cancer-Associated Fibroblasts Using Morphodynamic and Motile Features

Abstract

The heterogeneous natures of cancer-associated fibroblasts (CAFs) play critical roles in cancer progression, with some promoting tumor growth while others inhibit it. To utilize CAFs as a target for cancer treatment, issues with subtypes of CAFs must be resolved such that specific pro-tumorigenic subtypes can be suppressed or reprogrammed into anti-tumorigenic ones. Currently, single-cell RNA sequencing (scRNA-Seq) is a prevalent strategy for classifying CAFs, primarily based on their biomolecular features. Alternatively, this study proposes assessing CAFs on a larger biophysical scale, focusing on cell morphological and motile features. Since these features are downstream effectors of differential gene expression combinations, they can serve as holistic descriptors for CAFs, offering a complementary strategy for classifying CAF subtypes. Here, we propose an artificial intelligence (AI) classification framework to comprehensively characterize CAF subtypes using morphodynamic and motile features. This framework extracts these features from label-free live-cell imaging data of CAFs employing advanced deep learning and machine learning algorithms. The results of this study highlight the ability of morphodynamic and motile features to complement biomolecular features in accurately reflecting CAF subtype characteristics. In essence, our AI-based classification framework not only provides valuable insights into CAF biology but also introduces a novel approach for comprehensively describing and targeting heterogeneous CAF subtypes based on biophysical features.