NCG enrichment implemented in DOSE

I would like to give a big thanks to Guangchuang Yu, the author of many cool R libraries like GoSemSim and ggtree, for implementing a Network of Cancer Genes enrichment function in the DOSE R library.

The new function is called enrichNCG, and can be found in the github version of DOSE. You can use it to analyze a list of genes, and determine if they are enriched in genes known to be mutated in a given cancer type. For example, a random list composed by genes having an Entrez Id between 4000 and 9000 is enriched in genes mutated in sarcoma and leukemia:

If you have multiple sets of genes, you can also use the clusterProfiler library to compare them at the same time. Read this previous post for more examples of this functionality.

If you also have gene scores (e.g. a value for the expression or conservation of each gene), you can do a Gene Set Enrichment Analysis, which will give more importance to genes with higher scores:

You can also produce many nice plots. For example this is a cnetplot, in which each gene is connected to the terms related to it:

It is worth to mention that the DOSE package allows to calculate enrichment in the Disease Ontology database, which associates genes to disease terms. In my experience, for bioinformaticians Disease Ontology is more useful than OMIM, because it provides a clear association between genes and disease terms. If you use the raw OMIM data instead, you will have to text mine the descriptions and that can lead to a lot of noisy data.

Have a good enrichment with DOSE and NCG 😉

Solidarity to the workers of the Mario Negri Sud institute

I don’t usually post online petitions, but this is a bit personal to me as it regards the closure of a big research institute close to my hometown in Italy.

The Mario Negri Sud was a research institute active for the last 30 years in Abruzzo, a region in the center-south of Italy. During all these years the institute achieved excellence in many fields, from cardiovascular diseases to breast tumors, from diabetes to some rare diseases, and much more. I remember reading about an article on polycythemia vera (a rare cancer) published on NEJM just before financial problems halted most of the research, and more work by a neuroscientist who was really affected by the stress of the situation.

Unfortunately last week, after 4 years of financial struggle, the institute was officially declared bankrupt. Still, this is not the worst part.  Given the financial situation it is likely that the workers of the institute, aproximately 160 people between researchers and staff, will not receive their pay from the last 18 months, moreover they will not get their pension funds (the Italian TFR), which for some people amounts to tens of thousands of euros, accumulated in more than 25 years of work.

Negri sud picture
The Negri Sud institute. Click here to sign the petition.

These are people who dedicated almost all their life to research, and it is very unfair that they are threated this way. It is well known that the life of researchers is full of sacrifices and is never financially stable. To think that after many years they are denied a pension and abandoned to their fate is really inconceivable. Politicians were not able to solve the situation, and they are probably guilty of making it worse. Moreover given the geographical isolation of the institute, this situation hasn’t received much attention from the media outside of Abruzzo.

If you want to sign the petition, just click on this link to change.org. The petition is in Italian, and basically asks to the presidents of the Mario Negri institute, the Abruzzo region and the Chieti province (the three founding entitites of Negri Sud), to at least pay the pension and the salary of these workers. The change.org website will ask for your name, direction, postal code, and email. The website may later send you additional emails regarding other petitions, but you can opt out at any time.

Updates on docker and bioinformatics

My previous post on docker and bioinformatics received some good attention on Twitter. It’s nice because it means that this technology is getting the right attention in the bioinformatics community.

Here are a few resources and articles I’ve found thanks to the conversations in Twitter.

  • Performances of Docker on a HPC cluster – a nice article showing that running a NGS pipeline in a docker container costs about 4% of the performances. It’s up to you to decide whether this is a big or a small price to pay.
  • biodocker is a project by Hexabio aiming at providing many containers with bioinformatics application. For example, you can get a virtual machine with biopython or samtools installed in a few minutes. Update: this may have been merged with bioboxes (see discussion)
  • oswitch is a nice implementation of docker from the Queen Mary University of London, which allows to quickly switch between docker images. I like the examples in which they run a command from a virtual image and then return directly to another environment.
  • ngeasy, a Next Generation Sequencing pipeline implemented on Docker, by a group from the King’s College of London (I work in the same institute but I didn’t know them!).
  • a nice discussion on Biostar on how a reproducibility problem could be solved with Docker.
  • a Docker symposium planned for the end of 2015 here at King’s.
  • BioPython containers by Tiago Antao, including some ipython tutorials

Docker is another innovation for data analysis introduced in 2014. I am surprised by how many good things were released last year, including docker and the whole dplyr/tidyr bundle. Let’s see what 2015 will bring!

Reproducible bioinformatics pipelines with docker

I have recently come across a nice article explaining what Docker is and how it can be useful for bioinformatics. I’ll leave you to the article for more details, but basically Docker is an easy way to define a virtual machine, which makes it very straightforward for other people to reproduce the results of an analysis, with little effort from our side.

For example, let’s imagine that we are just about to submit a paper, and that our main results are based on the Tajima’s D index from the data in 1000 Genomes. The journal may ask us to show how to reproduce the analysis: which files did we used as input? Which tool did we use to calculate the Tajima’s D?

In this case, a docker file may be like the following:

The first part of this docker file will set up an ubuntu virtual machine, and install all the software needed to execute the pipeline: tabix, vcftools, snakemake. The second part will  clone the latest version of the pipeline in the virtual machine, and then use tabix to download a portion of chromosome 22 from the 1000Genomes ftp. The third part runs the pipeline, by executing a snakemake rule.

You can run this docker container by running the following:

This will take quite a while to run, and will build a docker virtual image in your system. Afterwards, you can run the following:

This command will open an interactive shell in the virtual machine. From there you will be able to inspect the output of the pipeline, and eventually, if this pipeline were more complex than a mock example, run other rules and commands.

This system makes it very easy to provide an environment in which our results can be reproduced. It is also very useful if we work from more than one workstation – e.g. if we need to have the same configuration at home and in the lab.

Just a few more links on docker and bioinformatics:

the most useful R function of the week: unnest from tidyr

There are many great functions in CRAN and BioConductor, and certainly saying that unnest from the tidyr package is the best is a big exaggeration. However this function solved a big problem in data formatting that made me waste a lot of time in the past, that I was surprised no one had implemented a function for it yet.

Imagine we have a dataframe like the following:

The first column contains the Entrez of each gene. This columns is fine, as it contains only one value per row, and it is easy to query or join with other dataframes. The second column, however, contains a comma-separated list of gene names, all associated to the same Entrez IDs. This column is a mess to deal with, because we need to use grepl to query it, and we can’t join it with other dataframes as long as it is in this form.

The unnest function from tidyr allows to convert this data frame in a “tidier” form, containing one row for each combination gene symbol and alias:

This code makes use of the %>% and some functions from the dplyr package, but it is still R!

Having the dataframe in this long form makes it a lot easier to deal with it. For example, let’s imagine that somebody asks us to get the Entrez IDs for the list of gene symbols DER7 and DEC. We would just have to do a simple subset on the dataframe:

This is just a silly example, which may have been solved with some application of apply and grepl, but in the real world there are a lot of more complex applications for it. For example, here is some code I used to split Blat output into one line per exon (or blat alignment block):

Continue reading

a formula interface for GeneOntology analysis

clusterProfiler is a nice R library for doing GO and KEGG enrichment analysis. It has a simple interface and it can produce some clear plots using the ggplot engine. Today I contributed a formula interface to clusterProfiler, making it easier to do enrichment of multiple groups of genes.

Let’s imagine you have a dataframe in which one column contains a list of Entrez Ids, while the other columns encode some grouping variables:

The new formula interface allows to do a GO analysis on each of the groups. For example, we can group them  by the column “group”, and compare the classification of the two groups:

clusterProfiler example

In this case group A is enriched in membrane and extracellular region, while group B is only enriched in membrane genes. The groupGO function used here doesn’t provide p-values – we should have used enrichGO instead. I guess that the 3 Entrez ids in group A correspond to 5 genes in Gene Ontology, so that’s why the plots shows a total of 5 in group A. See clusterProfiler’s documentation for better examples.

The formula interface allows also multiple grouping. For example:

Of course this example is not much interesting, since it is only 6 randomly chosen genes. However with bigger datasets the formula interface can be much more powerful.

Now that I think of it, it would be better if compareCluster would return a dataframe with multiple columns, instead of merging them into a single column called Cluster. This would be make it possible to plot the results using facets or something more fancy. It would be something like this:

However this would probably require introduce some retro-incompatibility in the library, and it is not a big deal as the Cluster column can be easily split using the separate function from tidyr.

farewell to Barcelona

I am posting this a bit late (since I already moved 6 months ago); anyway, the news is that I left my lab in Barcelona, and moved to London!

prbb from avda aiguader
The institute where I did my PhD: the PRBB in Barcelona. On the other side of the building there is the beach.

I am satisfied about my time in Barcelona, where I did my master thesis and my PhD on network theory applied to human population genetics. However, it was time to move and try new experiences.

a picture taken from the 4th floor of the PRBB
a picture taken from the 4th floor of the PRBB

Apart from the change of city, I also changed my field of work, as I am now working on cancer genetics. My new group is a young group recently moved from Italy to London, famous for research on the systems-level properties of cancer genes , for a database called Network of Cancer Genes, and involved in a consortium for the sequencing of hepatocellular carcinoma. I will keep you informed of the proceedings!

my first PyPI package: vcf2networks

My first Python package is in PyPI!! I guess that now I can officially call myself a python programmer.

VCF2Networks is a python script to calculate genotype networks from population genetics data. Genotype networks are a method used in systems biology to study the “innovability” of a given phenotype, by representing all the genotypes associated with the phenotype as a graph, and studying some properties of this graph, such as the average path length and the average degree. For more info, you can look at the slides of the “Origins of Evolutionary Innovations” book club in this blog. The script in VCF2Networks allows to take any dataset of genotypes stored in the VCF format, and calculate many of these properties.

In principle, I am planning to submit an application note about the script to a bioinformatics-oriented journal. So, if you have some little time to lent me, and you want to test it, any feedback will be very useful for me. At the moment, the major issue is to simplify the installation, because this package depends on numpy and python-igraph, and these two modules require some terrible C libraries that must be installed separately. If you are aware of any way to distribute a binary package of a python module that depends on C libraries, your suggestion will be really welcome.

The presentation of my PhD defence

That’s it! Last week I defended my PhD thesis!! I have gone through it, and survived to tell!

I don’t feel very different from before, apart from being relieved :-). Now the future is possibly more difficult than before, because I have to look for a job position and finish a lot of things.

While I was preparing the slideshow, I realized that there are not many examples of presentations for a PhD defence online. This is bad, because you need all forms of help to prepare this presentation.The PhD defence is the last thing that you do as a PhD student, so you want to do it perfectly. It is also the moment when you describe many years of your work to the your colleagues and family. Thus, it is bad that there are few examples of slideshows for PhD defence online.

Here is the presentation that I have prepared for my defence. I hope that it will be useful to other people as an example for their defences.

I think that, for this type of presentation, the first slide to make is the “summary of the talk” slide, like the “Topics” slide I have. Usually I don’t like to have such summary slides in my presentation, but for the Thesis defence it is very important, because it gives you a feeling of security when you present. Having a well defined structure allows you to know when you can stop to drink some water or to check if everybody is following, and to know exactly what to say in each slide of the talk.

my poster featured in the “Better Posters” blog!

My ECCB2012 poster has been featured in the Better Posters blog. Check the article here: http://betterposters.blogspot.com.es/2013/10/invitating-interaction.html

I am glad because betterposters is one of my favorite blogs. It’s a blog about designing and improving posters for scientific conferences, and it contain many tips and examples of how scientific posters can be improved.

The poster featured there is the poster of the “Post-its”, which I briefly described in the article of the “best practices“.

DSC00132
Here are some other comments and tips from my experience of using post-its to get feedback during a conference:

Continue reading