|
|
| |
1. |
Introduction of fmpRPMF Web Server [+] |
|
|
The fmpRPMF web server was developed for the identification of protein by Feature Matching Pattern based SVM approach.
fmpRPMF focused on the inherent attributes and critical issues of theoretical spectrum (peptides), experimental spectrum (peaks) and spectrum (masses) alignment. Eighty-one feature matching patterns derived from cleavage type, uniqueness and mass altering of theoretical peptides together with intensity rank of experimental peaks were firstly proposed to characterize the matching profile of peptide mass fingerprinting procedure. With the participation of matched peak intensity redistribution, a new strategy developed to handle shared peak intensities, four hundred and forty parameters were generated to digitalize each feature matching pattern. A high performance for evaluation dataset of 137 items was finally achieved by the optimal multi-criteria SVM approach with 491 final features out of a feature vector of 35640 normalized features through cross training and validating publicly available gold standard PMF dataset of 1733 items.
There are two stages for the fmpRPMF web server. In the first stage, 21 features out of 491 fmpRPMF features were selected to crude rank candidate proteins from first scanning. Crude ranking can reduce the number of candidate proteins from thousands to tens. In the second stage, top crude ranked candidate proteins were subscribe to SVM process to be predicted with positive and negative probability.
|
|
|
| |
2. |
|
|
|
Peak intensity, which has been ignored by traditional methods due to its variety derived from various ionization processes and unequal tryptic specificity for different amino acids, is now becoming to a hot issue. It is supposed that intense experimental peaks are more likely to be generated from true peptides rather than noisy, which has been proofed by previous findings together with successful prediction about relationship between peak intensity and protein sequence.
Hence, fmpRPMF gave prominence to peak intensity in this approach by using it rather than precisely predicting it. Firstly, peak intensity proportion was newly applied to make it more comparable between different PMF data. Secondly, peak intensity and peak intensity proportion have participated in the generation of all series of parameter. Thirdly, parameter series of top matching have been newly proposed to highlight the effect of intense experimental peaks in the performance. Finally, parameters derived from peak intensity and peak intensity proportion contributed 5 out of 11to the selected SVM parameters. As an inherent attribute of experimental spectrum, peak intensity should be naturally paid more attention during the protein identification.
It’s very easy to generate PMI file. Firstly, extract masses plus Relative Intensities list from experimental spectrum by user’s own MZ software. Secondly, make sure that masses are arrayed in first column and mass relative intensities are arrayed in second column oppositely. Two columns are separated by space or tab. It’s not necessary to sort masses or mass intensities. Lastly, save data as plaint format with suffix ".pmi". The demo of builting pmi file is shown as below.
The relative intensities created by MS software is recommended for the fmpRPMF server. However,fmpRPMF server can handle absolute intesity value such as peak height and peak area. The sever converts absolute value to relative value automatically during the upload process. Please see the demo pmi file spf27_human.pmi.
|
|
|
| |
3. |
fmpRPMF Searching Parameters [+] |
|
|
fmpRPMF has a mechanism named MPIR to handle the multiple shared peak intensities during spectrum alignment. Differentiated weights make it rational to allocate shared peak intensities to various originated peptides. Meanwhile, MPIR has the inherent mechanism to take both peptide missed cleavage and mass modification into consideration concurrently. The tuning of MPIR showed that missed cleavages should be no longer obstacle to protein identification. Normalized by several types of harmonic factors, missed cleavages and mass modification can offer competitive information to the performance. Recommended searching parameters as below:
Recommended Searching Parameters |
Database |
PUD derived from UniProtKB/Swiss-Prot |
Enzyme |
Trypsin |
Allow up to |
1 missed cleavages |
Fixed modifications |
Carbamidomethylation of C |
Variable modifications |
Oxidation of M |
Peptide tol. ± |
100ppm |
Mass values |
MH+ |
Monoisotopic |
Average |
Report top |
AUTO hits |
All optional and interactive parameters can be adjusted in the web page before submiting.
|
|
|
| |
4. |
PUD(Peptide Uniquness Database) [+] |
|
|
For each theoretical digested peptide, an implied property is the repeat number of its sequence in the whole protein sequence database. A peptide could be assigned as unique peptide if it exists in only one protein. A local secondary database PUD (Peptide Uniqueness Database) derived from SwissProt has been constructed to generate tsPUN for each peptide.
Biweekly updates of PUD are synchronized from Swiss-Prot by anonymous FTP (ftp://ftp.expasy.org/databases/swiss-prot/updates_compressed/) |
|
|
| |
5. |
|
|
|
2016.10.15 Update the crude ranking process
2015.07.08 Update Perl and Mysql to 64bit version
2014.01.22 Optimized the SVM predicting process
2012.11.21 Add flag counter for web visiting
2012.09.20 More optional and interactive parameters are applicable for web server
2012.02.11 Auotmatically run sample pmi
2011.05.31 Modification of HTML Javascript to compatible with Firefox
2011.04.15 Add stastistics of web visiting
2010.11.12 Beta version |
|
|
| |
6. |
|
|
|
- Youyuan Li, Pei Hao, Siliang Zhang, Yixue Li.Feature-matching Pattern-based Support Vector Machines for Robust Peptide Mass Fingerprinting.Molecular & Cellular Proteomics,doi: 10.1074/mcp.M110.005785
- Youyuan Li, Yingping Zhuang. fmpRPMF: A Web Implementation for Protein Identification by Robust Peptide Mass Fingerprinting.IEEE/ACM Transactions on Computational Biology and Bioinformatics,doi:10.1109/TCBB.2017.2762682
|
|
|
| |
7. |
|
|
|
Any advice or feedback please contact leeyy77@gmail.com |
|
|
|
|
Feature-matching Pattern-based SVMs Approach for Robust Peptide Mass Fingerprinting