Statistical models to characterize colon tumor stiffness heterogeneity through representative atomic force microscopy maps.

The study of the impact of physical forces and on cells has emerged as a fertile field of investigation. Applications in oncology are especially transformative since tumor stiffness was found associated with fundamental processes such as tumorigenicity, disease progression, and resistance to therapy. We present an integrated atomic force microscopy-statistical modeling and machine learning approach in colon cancer to link clinical and phenotypical parameters with local maps of tissue stiffness. Statistical modeling found both known associations, such as with age or tumor stage, but also novel associations, such as with RAS mutations, tumor left-/right-colon localization, and DNA repair deficiencies (microsatellite instabilities). Machine learning was able to infer clinical parameters from stiffness data. This work establishes a computational framework to build global models integrating all the available clinical parameters and assess their relevance with respect to stiffness, while they are usually explored separately. Similarly, we established spatial statistics techniques to interpolate and model the topographical information embedded in local stiffness maps.