Motivation: Baseline removal, as the first preprocessing step of SELDI data, critically influences subsequent analysis steps. Current baseline removal algorithms of SELDI data, which are based on mathematical morphology, result in biased signal estimates. Due to the parameterization of current algorithms for baseline removal, noise and spectral signal distributions bias the removal results, which may lead to seemingly interesting but ultimately irreproducible results in downstream analysis.

Results: We proposed a Heuristic Based Baseline Removal (HbBr) algorithm to model the baseline. HbBr first identifies the potential peak regions by utilizing first derivatives and amplitudes information as a fast heuristic, then down-weights peak regions before modeling the baseline with a nonparametric smooth curve. It outperformed mathematical morphology-based algorithms implemented in the PROcess package of Bioconductor, as judged by a series of benchmark experimental data sets and simulated data sets. We also found that HbBr is computationally more efficient than PROcess. Furthermore, we demonstrated that the HbBr algorithm, although designed for SELDI, yields a good baseline correction of MALDI data without adjusting any parameters.
  

mass spectrometry, baseline, proteomics