Introduction to Feature selection for bioinformaticians using R, correlation matrix filters, PCA & backward selection

Bioinformatics is becoming more and more a Data Mining field. Every passing day, Genomics and Proteomics yield bucketloads of multivariate data (genes, proteins, DNA, identified peptides, structures), and every one of these biological data units is described by a number of features: length, physicochemical properties, scores, etc. Careful consideration of which features to select when trying to reduce the dimensionality of a specific dataset is, therefore, critical if one wishes to analyze and understand their impact on a model, or to identify what attributes produce a specific biological effect.

For instance, considering a predictive model C1A1 + C2A2 + C3A3 … CnAn = S, where Ci are constants, Ai are features or attributes and S is the predictor output (retention time, toxicity, score, etc). It is essential to identify which of those features (A1, A2 and A3…An) are most relevant to the model and to understand how they correlate with S, as working with such a subset will enable the researcher to discard a lot of irrelevant and redundant information.

What is your tool for peptide/protein identification?

In 2012 We published a Poll about most used softwares from peptide/protein identification in proteomics in Computational Proteomics Linkedin Group. I decided to reproduce the Poll here because linkedin remove this option and also here i have the opportunity to add more softwares to the list.

My List of Most Influential Authors in Computational Proteomics (according to Articles References, Google Scholar, twitter, Linkedin, Microsoft Academic Search and ResearchGate)

Young researchers starting their careers will often look for reviews, opinions and research manuscripts from the most influential authors of their chosen field. In science, however, unlike many other topics on the Internet, ranked lists or manuscript repositories of top authors sorted by research topic are hard to come by. For some researchers, the idea of such a task brings the words ‘wasted time’ to their minds; the most critical condemn it as a frivolous pursuit. Maybe so. In my opinion, however, it as an excellent starting point.

Home Page of ResearchGate with more than 3 millions of users

These days, more people than ever are involved in science and research. Just look at ResearchGate’s homepage.  There are over 3 million persons there –and we’re only counting ResearchGate users. Once simple undertakings, such as finding the right manuscript to cite, the most authoritative group on a topic, or the best software application for a specific task, have become increasingly difficult for graduate students navigating this ocean of data, despite the availability of services such as Google Scholar or Pubmed. The situation will only worsen in the future, as is easy to see by simply tallying the number of  published papers in the fields of Proteomics, Genomics, Bioinformatics and Computational Proteomics since 1997:

Number of published manuscripts in Pubmed per year (1997-2012). the statistics was done using the Medline Trend Service http://dan.corlan.net/medline-trend.html

In 2012 alone, over 6,000 and 17,000 manuscripts were published in the fields of proteomics and bioinformatics, respectively. Our young field, computational proteomics, published more than four hundred papers. Perhaps well-established PI’s or Group Leaders can easily tell apart derivative or me-too contributions from groundbreaking work, but young scientists, who spend most of their time implementing someone else’s ideas, can certainly have a hard time doing so. Although technology has come to the rescue with today’s mixture of search engines and social networking tools (ResearchGate, Google Scholar, twitter and LinkedIn among them), the best way to harness its power is, precisely, by starting from a ranked list of the most authoritative voices within a field of research, whose whereabouts can then be traced in the scientific literature, the blogosphere, and anywhere else.

Celebrating Ten Years of Mann and Aebersold’s “Mass spectrometry-based proteomics” review.

In 2003 Mann & Aebersold reviewed on the pages of Nature the challenges and perspectives of the then-nascent field of MS-based proteomics. Mass spectrometry (MS) has since entrenched itself as the method of choice for analyzing complex protein samples, and MS-based proteomics has become an indispensable technology for interpreting genomic data and performing protein analyses (primary sequence, post-translational modifications (PTMs) or protein–protein interactions).

" The ability of mass spectrometry to identify and, increasingly, to precisely quantify thousands of proteins from complex samples can be expected to impact broadly on biology and medicine."

The manuscript by Mann & Aebersold is one of the most cited manuscripts in the field of MS proteomics, For this reason is one of the “core papers” in the field of proteomics and computational proteomics, outlining most of the basic concepts required to understand the fundamentals of this discipline.

Ten years after its publication the main workflow described in the manuscript do not change dramatically. In this period major advances are related to the development of the Thermo’s Orbitrap Mass Spectrometer (Velos, LTQ, Exactive, etc) and new fragmentations types (ETD, HCD). Separation techniques (electrophoretic and chromatographic) were explored extensively in these ten years. Aebersold pioneered in 2005 the use of OFFGEL electrophoresis and electrophoresis fragmentation at peptide level (Heller 2005) and Mann’s group developed the FASP method for sample preparation before protein digestion (Wiśniewski JR et al 2009),both of which have contributed significantly to the dramatic increase in the number of identified proteins characterizing today’s proteomic projects. Surprisingly, the development of electrophoretic methods in the last 3 years looks like a “passed-on topic”. In ten years we moved from identifying at most 500 species in complex samples to identifying 60% of the human proteome.

Retrieve weka weitghts from Linear Regression

Weka function to retrieve the linear regression weights:

import java.util.List;
import weka.classifiers.Evaluation;
import weka.classifiers.functions.SMOreg;
import weka.classifiers.functions.supportVector.Kernel
.....

public double[] getCoefficients(){
       String svmClassifier = svm.toString();
       List<Double> coefficents = new ArrayList<Double>();
       double[] coeficcientsArray = null;
       if(svmClassifier.contains("weights (not support vectors):")){
          String[] svmClassStrings = svmClassifier.split(System.lineSeparator());
          int i = 0;
          boolean lastFound = false;
          while( i < svmClassStrings.length && !lastFound){
              String currentString = svmClassStrings[i];
              if(currentString.contains("weights (not support vectors):")){
                  i++;
                  while(i < svmClassStrings.length && !lastFound){
                      currentString = svmClassStrings[i];
                      String[] parts = currentString.split("\\s+");
                      if(parts.length < 5){
                          lastFound = true;
                      }else{
                          String number = parts[1] + parts[2];
                          coefficents.add(Double.parseDouble(number));
                      }
                      i++;
                  }
              }
              i++;
          }
        } 
 

#HavanaBioinfo2012 Hard, but Awesome Experience

From 8th to 11th of last December the "I Bioinformatics for Biotechnology Applications" (#HavanaBioinfo2012) was held in the hotel “Occidental Miramar” located within an elegant area in Havana, Cuba. Putting on a Bioinformatic workshop takes a lot of different pieces (small ones and big ones). You need write lot of mails to invited speakers, you need a nice and comfortable place with lots of tables and chairs and good food and the drinks. 

I started from October, writing the first mails to EBI friends, advisers and professors (www.ebi.ac.uk), and to be complete honest it was fantastic because most if then accepted from the very beginning. Thanks to Henning Hermjakob and Alex Bateman (@Alexbateman1) for the support. At the end the workshop was fully subscribed, with more than 45 attendees and sixteen speakers, participating in two poster sessions and a panel discussion. Speakers from EBI, Belgium University, Mascot (UK), Bioinformatics Solutions (Canada) accepted the invitation to come here and give one or two lectures about proteomics, genomics, bioinformatics.