BIOINFORMATICS APPROACHES TO MALDI-TOF MASS SPECTROMETRY DATA ANALYSIS

Despite the increasing performance of Mass spectrometry (MS) and others analytical tools, only few biomarkers have been validated and proved to be robust and clinically relevant; indeed a large numbers of proteomic biomarkers have been described, but they are not yet clinical implemented [1]. MALDI-TOF MS seems one of the more powerful tool for biomarkers discovery [2, 3], and shows interesting clinical properties, for instance the possibility to directly search in peripheral fuids for proteins related to an altered physiological state: samples (urine, plasma, serum, etc.) can be collected easily and cheaply by non-invasive, or very low-invasive, methods [4]. The combination of some biomarkers is actually considered more informative than a single biomarker [5, 6], and the improvement in the bioinformatics analysis of MS data could probably help this investigation, decreasing costs and time necessary for each discovery [7]. It is possible to approach the problems related to the analysis of (MALDI-TOF) MS data in two ways, either trying to increase the number of available samples or by reducing the complexity of the problem [8]: in the first case, we developed an approach to compare small datasets from different sources (i.e. hospitals), based on mutual information and mass spectra alignment, that showed significant performance increase compare to the competing ones tested. In the latter case, we developed novel methods and approaches to compare MALDI-TOF MS profiles of normal and Renal Cell Carcinoma (RCC) patients, with the goal of isolating the more interesting subset of small proteins and peptides from the whole analysed peptidome. MS-based profiling is in fact able to detect differently expressed proteins or peptides during physiological and pathological processes. Every MALDI-TOF MS spectrum, that reports the relative abundance of sample analytes, could be considered as a snapshot of samples peptidome in a definite mass range. The relationship between mass/charge ratio, or m/z, and concentration of detected peptides can be represented by networks. Tumor case and control subjects show different peptidome profiles, due to differences in biomolecular and/or biochemical features of cancer cells: they will show some changes in the networks that describe them. We use graphs to create networks representation of data and to evaluate networks properties. We explore the networks properties comparing cases versus controls datasets, and subdividing cases in the different histological subtypes of RCC, clear cell RCC (ccRCC) and not-ccRCC, using different methods both for networks creation and analysis, and for results evaluation. We identify, for each datasets (controls, ccRCC and not-ccRCC) some interesting mass ranges within which we believe biomarkers signals should be searched. In conclusion, we have developed a set of methods which we believe improve the current computational approaches for the analysis of mass spectrometry data. These results have been published or presented at workshops and conferences.