Fig. 2.

The schema illustrates basic steps of the MSClust algorithm. A—computing PD of each ion fragment based on two distances: the retention time distance between mass peak peak apices (determined by an alignment software) (X-axis of A, B and C) and an intensity pattern similarity distance (Y-axis of A, B and C). The more close neighbours an mass peak has in the two-dimensional feature space, the higher its PD is (the darker its dot in plot A). B—selection of ‘centrotype’ ion fragments as centres of clusters (cA and cB). C—classification: computing memberships of each ion fragment in the cluster centers. The dots depicted in brown have uncertain (intermediate) membership and can represent mass peaks common for cA and cB. D—conversion of clustering results into reconstructed mass spectra (‘ms A’ and ‘ms B’) and selection of most representative mass peaks (‘qA’ and ‘qB’). The red-green color scale below reflects the membership of mass peaks in cluster A (green) and B (red)