Machine learning identifies remodeling patterns in human lung extracellular matrix

Abstract

Organ function depends on the three-dimensional integrity of the extracellular matrix (ECM). The structure resulting from the location and association of ECM components is a central regulator of cell behavior, but a dearth of matrix-specific analysis keeps it unresolved. Here, we deploy a high-resolution, 3D ECM mapping method and design a machine-learning powered pipeline to detect and characterize ECM architecture during health and disease. We deploy these tools in the human lung, an organ heavily dependent on ECM structure that can host diseases with different histopathologies. We analyzed segments from healthy, emphysema, usual interstitial pneumonia, sarcoidosis, and COVID-19 patients, and produced a remodeling signature per disease and a health/disease probability map from which we inferred the architecture of healthy and diseased ECM. Our methods demonstrate that exaggerated matrix deposition, or fibrosis, is not a single phenomenon, but a series of disease-specific alterations.

Statement of significance

The extracellular matrix, or ECM, is the foremost biomaterial. It shapes and supports all tissues while regulating all cells. ECM structure is intricate, yet precise: each organ, at every stage, has a specific ECM structure. During disease, tissues suffer from structural changes that accelerate and perpetuate illness by dysregulating cells. Both healthy and diseased ECM structures are of great biomedical importance, but surprisingly, they have not been mapped in detail. Here, we present a method that combines tissue engineering with machine learning to reveal, map and analyze ECM structures, applied it to pulmonary diseases that kill millions every year. This method can bring objectivity and a higher degree of confidence into the diagnosis of pulmonary disease. In addition the amount of tissue needed for a firm diagnosis may be much smaller than required for manual microscopy evaluation.