Identification of PNH Affected Cells by Classifying Motion Characteristics of Single Molecules

In this paper we describe an approach for identifying and classifying motion characteristics of single molecules for providing the evidence of paroxysmal nocturnal hemoglobinuria (PNH) in nano-scale microscopy images. The main goal is to define appropriate features of trajectories of single molecules and then to identify classifiers that are able to correctly distinguish between PNH affected and healthy cells using these motion characteristics of single molecules. First, single molecule detection in microscopy images is performed. Afterwards, trajectories of single molecules are identified using a nearest-neighbour algorithm; the so determined trajectories are analysed on the basis of diffusion constants. A set of 9 features is calculated for all trajectories; this information is in combination with class labels (healthy vs. PNH affected) used as input for machine learning algorithms such as the k-nearest-neighbour algorithm, support vector machines, random forests, and genetic programming. Using the implementations of these algorithms in HeuristicLab 3.3.8, up to 82.87% of the here analysed trajectories can be classified correctly.