Joint corresponding feature identification and alignment for multiple LC/MS replicates

In Liquid Chromatography/Mass Spectrometry (LC-MS), identifying corresponding peptide features (LC peaks) in multiple replicate datasets plays a crucial role in the differential analysis of complex peptide or protein samples for biomarker discovery. Given a peptide sequence, we aim at identifying its LC peak intervals in all datasets simultaneously. Generally, features are first identified in each replicate dataset, and then the features are aligned using warping functions. In such a procedure, the error in feature identification will propagate to alignment. Instead, we consider the problem of joint feature identification and alignment in multiple datasets. Since accurate feature identification improves the accuracy of corresponding feature alignment and vice versa, joint processing provides better performance than separate processing. We propose an algorithm which combines peak identification quality scores, time shifts and the similarity of LC peak shapes between candidate corresponding features for accurate alignment. In addition, we also incorporate the approximate elution time interval of a peptide stored in an Accurate Time and Mass (ATM) database when available. We test our algorithm on publicly available datasets, and we compare its with that of separate feature identification and alignment. Results show that the number of accurately identified corresponding features is improved significantly by using the proposed method.