The invention relates principally to the statistical analysis of protein separation patterns. The invention solves the problems associated with producing models which are predictive of classification using unreduced data. The invention provides a method of analysing a representation of a separation pattern, the representation including a neighborhood representing a region of the separation pattern, the neighborhood including a plurality of data points, the method comprising augmenting data by representing the entire region using each data point of the neighborhood; and building a classification model using some or all of the data points.