Application of ensemble machine learning methods to multidimensional AFM data sets

摘要

Multidimensional data sets collected with atomic force microscopy on complex biological objects like cells or tissues could be extremely informative. However, due to multidimensionality and unavailability of a large number of samples, processing of such data could be a challenge for automated machine learning methods. Here we discuss an approach based on a reduction of data dimensionality when only a limited number of parameters calculated from each microscopy map are used for machine learning algorithms. This method requires a smaller number of imaged cells, demonstrates higher accuracy of prediction, and provides cell identification that is independent of operator involvement.