Proteinmarker Detection Software™ (PDS™)

PDS™ is part of the MBS™ bioinformatics package!

PDS™ is a stand-alone software tool that performs analysis on Matrix-Assisted Laser Desorption/Ionization Time-Of-Flight Mass Spectrometry (MALDI-TOF MS) spectra and Surface-Enhanced Laser Desorption/Ionization Time-Of-Flight Mass Spectrometry (SELDI-TOF MS) spectra. PDS™ performs all analyses required for a complete proteomics mass spectrometry data analysis study. No other software are required and tools such like Matlab, R or Excel no longer needs to be used for standard data processing.

The quality of analysis for PDS™ is very high. The software also allows all results to be exported and used in your own self-developed software tools. PDS™ is compatible with most MALDI and SELDI mass spectrometers. PDS™ also speeds up your analysis process allowing you to save time and costs.

PDS™ Helpfiles

STEP 1 - How to run a quick analysis using PDS (2,01 MB)

STEP 2 - How to export analyzed SELDI and MALDI spectra from PDS (788 kB)

STEP 3 - How to use Binning (232 kB)

STEP 4 - How to change important settings (572 kB)

STEP 5 - How to use PDS for non-SELDI or non-MALDI spectra (184 kB)

PDS: Help file (5,66 MB)

PDS™ SELDI-TOF MS Datasets

SELDI spectra / dataset (2,58 MB)

Module 1: Pre-Extraction Processing

Sub-Module 1: Calibration

Sub-Module 1 uses a calibration algorithm to transform one-dimensional time-of-flight data into two-dimensional spectra with intensities values on the y-axis and mass-through-change (M/Z) values on the x-axis. This method is automatic and it is only activated for SELDI data when one-dimensional SELDI time-of-flight data are detected by the software during the importing of a dataset. Calibration is not used for standard two-dimensional (Intensity / mass-through-change) MALDI or SELDI data.

The Calibration algorithm is based on Ciphergen's conversion of time-of-flight (t) data to mass (m/z) values via a quadratic equation,

...where U is the known ion voltage (20 000 for these data) and sign(t -t0) = 1 for t > t0 and -1 otherwise. For more information about Calibration, please read the following paper by Neal Jeffries:

Sub-Module 2: Baseline Subtraction

Sub-Module 2 allows you to perform baseline subtraction on spectra. The analysis calculates and removes the baseline signal from each spectrum. Two options are available; 1) calculating the baseline beneath (but along with) all the bottom data points, and 2) calculating the average baseline between the noise area placing half of the noise area above the baseline and half below the baseline (peaks are excluded from the calculation).

Sub-Module 3: Total Ion Current (TIC) Normalization

Sub-Module 3 allows you to normalize your data by TIC (peak intensity/TIC). The analysis calculates either the total intensity in or the area under the curve of a spectrum, divides each intensity value in the spectrum with the calculated value, and multiplies it with a specified factor. The sum of all intensities will be equal to the selected value, and the value of 1 (100%) is set as default.

Module 2: Peak Extraction

Sub-Module 1: Peak Detection by Clustering

Peak Detection by Clustering is a unique peak detection algorithm developed by MedicWave.

Peak Detection by Clustering begins by calculating the baseline and noise level of each spectrum. The correct intensity of a peak is not the intensity measured by the machine, but the intensity measured by the machine after subtraction of the baseline. The software performs baseline subtraction and removes the baseline signal from each spectrum.

The noise level is an estimation of the noise contribution to the intensity. The noise level is an intensity level above the baseline and is not constant over an entire spectrum. The quotient between the intensity of a peak (after baseline subtraction) and the noise level (after baseline subtraction) is called the Signal/Noise Ratio, SNR. Peaks below the noise level (SNR < 1) are considered probable noise, whereas peaks above the noise level (SNR > 1) are considered probable peaks. Peaks with high SNR are more likely to be real peaks whereas peaks with low SNR are more likely to be noise. The noise level itself corresponds to SNR = 1. You can specify your preferred SNR in the software settings when you start the Peak Extraction analysis.

All peaks with an intensity value above the selected SNR are used to construct clusters. A cluster is defined as a consecutive sequence of datapoints. Peaks are calculated using a unique (by MedicWave developed) centroid calculation and clustering algorithm. Peaks below the selected SNR are considered to be noise and they are not used for futher analysis.

Peak Detection by Clustering is a very fast and robust peak detection algorithm, and it outperforms most peak detection methods available on the market. It is an invaluable tool for peak detection in Mass Spectrometry Proteomics.

Sub-Module 2: Peak Detection by Resolving Overlapped Peaks (ROP)

Peak Detection by ROP was developed in collaboration with Doctor Martijn Dijkstra from Groningen Bioinformatics Center (GBiC), University of Groningen, The Netherlands.

Peak Detection by Resolving Overlapped Peaks is a new procedure to decompose a spectrum into protein peaks and a baseline using so-called statistical finite mixture models. Major advantages of the new approach are (i) the detected abundance of a protein is quantified properly by estimating its area under the peak curve, (ii) areas under the curve can be estimated even if adjacent peaks overlap considerably, (iii) mass-resolution and abundance estimation of a protein are therefore substantially improved in such cases.

Resolving Overlapped Peaks begins by smoothing each spectrum to remove high frequencies. Only peaks above the SNR level are extracted while the rest are automatically discarded.

Two smoothing methods are available:

  • Savitzky-Golay filter
  • FFT filter (high frequencies are removed)

A robust version of the EM-algorithm is applied to detect overlapped and hidden peaks.

Sub-Module 3: Peak Detection by using ProSpect

Peak Detection by using ProSpect was developed in collaboration with Professor Yudi Pawitan, Doctor Chuen Seng Tan, Doctor Alexander Ploner, and Doctor Andreas Quandt from the department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm, Sweden.

ProSpect has a clear advantage for peak detection because it uses all spectra simultaneously to identify peak regions and it makes use of cross-spectral information.

ProSpect is a peak quantification algorithm combined with a signal detection algorithm which addresses low specificity and poor peak quantification of the standard peak detection method developed by Ciphergen. ProSpect outperforms the standard method in several aspects:

  • ProSpect captures several peak regions in the spectra that are missed by the standard method.
  • ProSpect is better at classifying regions as peak or non-peak regions.
  • ProSpect is less easily confused by two or more closely neighboring peaks and/or m/z-misalignment than the standard method.
  • ProSpect quantifies peaks much better than the standard method.

ProSpect is very similar to a method called SSA (Simultaneous Spectrum Analysis). SSA also performs a true multi-spectral approach based on (unmodified) F-statistics that outperforms the standard method in terms of peak detection and quantification. The crucial difference between the SSA and the ProSpect approach lies in the fact that the F-statistics in SSA are based on the biological grouping of the spectra, e.g. knock-out vs. normal mice or benign vs. cancerous prostate tissue. Although the ultimate interest is in the between-class variation, very few markers will be significantly different between classes, so if every region is tested across the range of spectra without any screening then the potential for false discoveries is large and it is harder to detect the real signal.

Publications concerning ProSpect

Module 4: Data Analysis

Sub-Module 1: Binning Analysis

Binning is the process of grouping measured data into data classes or histogram bins.

After binning peaks are no longer used. Only bin contents are used thereafter. Bins are often not of the same size, but they are homogenous between different data files. Data files after binning can therefore be compared with each other.

Binning can be performed using these two methods:

  • By calculating the distance between peaks
  • By calculating the density of peaks

Modules & Sub-Modules

PDS™ contains several Modules, and each of them contains several Sub-Modules. A Sub-Module can consist of one or several methods or algorithms.

Module 2, Sub-Module 1: Peak Detection by Clustering

Peak Detection by Clustering is a very fast and robust peak detection algorithm, and it outperforms most peak detection methods available on the market. It is an invaluable tool for peak detection in MALDI-TOF and SELDI-TOF Mass Spectrometry Proteomics.

Module 2, Sub-Module 3: Peak Detection by using ProSpect

ProSpect has a clear advantage for peak detection because it uses all spectra simul- taneously to identify peak regions and it makes use of cross-spectral information. ProSpect is a peak quantification algorithm combined with a signal detection algorithm which outperforms the standard SELDI-TOF MS peak detection method.