Fig. 1.

Illustrative example of the incorporation of grouping information into the similarity score. Each node in the figure is a peak feature, and dotted ovals represent groups of related peaks, e.g. isotopes, fragments, etc. Initially weights (e.g. W_AE) are computed for pairs of peaks (one from each run) with m/z and RT within pre-defined thresholds. These weights are converted into an overall score by incorporating grouping information. For example, peak pairs (A, E) and (B, G) are both within the threshold. As A and B are in the same group, and E and G are in the same group, the weights between pairs (A, E) and (B, G) are upweighted. Peak J is not related to any peaks that could be matched with A’s related peaks and the similarity between A and J is therefore downweighted (because $\alpha \le 1$). The same applies to similarities between pairs (C, H) and (D, I)