Choosing colors

While not specifically for R, ColorBrewer gives you an excellent reference for color range for different number of bins as well. Fantastic for heatmaps.

Machine learning (feature selection) in R

I have been taking a couple of machine learning classes on Coursera (Johns Hopkins - Practical Machine Learning, WashU - Introduction to Data Science, waiting for Stanford's course to be offered again!) and I find the following post to be very useful in explaining in detail the mechanics of feature selection.

http://topepo.github.io/caret/featureselection.html

It explains various methods and algorithms for feature selection in R. The caret package is particularly useful as it conveniently unifies different methods under a single function wrapper, train.

Big Data solutions in genomics

Are big data solutions for genomics all about increasing processing speed / handling it's size? Perhaps the most immediate application of big data solutions (or in most of these cases, Hadoop) is optimization or distributed computing. Examples of these are:

• Hadoop-BAM library for the scalable manipulation of aligned next-generation sequencing data in the Hadoop distributed computing framework
• CloudDOE a user-friendly tool for deploying Hadoop clouds and analyzing high-throughput sequencing data with MapReduce
• CloudBurst parallel read mapping 
• DistMap short read mapping on a Hadoop cluster
• SeqPig scalable scripting for large sequencing data sets in Hadoop
• Crossbow  implementation of exisiting tools (Bowtie, SoapSNP) to run on a parallel cloud cluster
• SeqAlto read alignment and resequencing
• Myrna Cloud-scale differential gene expression for RNA-seq

These are great examples of processing NGS data quicker and over a cluster/cloud/parallel system.

Can it do more? I would really like to see (and also develop myself if I can!) genomics solutions using these principles for:

• Statistical analysis
• Network inference
• Associations
• Metagenomics

Perhaps it is a good time to also talk about the application of non-relational databases (NoSQL) solutions. Genomic data is sparsely distributed (at least to me, in terms of storing metadata as well). For example, if you look at chromatin-bound entities (ChIP-sequencing), not all have the same bound sites. Similarly, a gene expression experiment will store data differently than a DNA mutation one. They do however share similar data, and more often than not, we have to combine them.

 I leave you with the videos of the Big Data in Biomedicine Conference 2014, which talks a lot on what kind of big data is being generated, what kind of computer technology is available to provide solutions, and a little bit at the end on the statistics and machine learning (best part!).

R: sequence logo from FASTA

Extracting fasta from bed in R

Extracting fasta from bed (specifying chromosome location)

Currently, bedtools is not readily available in R, but if you have bedtools installed (Mac/Unix) and it is in your path (export your path if not), you can write a simple function in R to replicate the commandline task.

Data Analysis in R : PCA

Good article on principle component analysis in R, how to do feature selection, etc. "Big data" is a hot topic at the moment, and it's relevance in bioinformatics cannot be ignored.

http://www.r-bloggers.com/introduction-to-feature-selection-for-bioinformaticians-using-r-correlation-matrix-filters-pca-backward-selection/

Coursera course review : Statistical Inference

Statistical Inference

This is a concise 4 week introductory course to basic statistics concepts, and how to perform statistical inference, offered by Johns Hopkins University. You will learn about sets, distributions, confidence intervals and multiple testing.

You do go through a fair bit of material in such a short time, and you would probably benefit with extra reading outside the course especially if you have not taken any statistics class before. The topics covered in this class is approximately the mixture of Biostatistics Bootcamp 1 and a small chunk from Biostatistics Bootcamp 2, both of which are 8 week courses offered by the same department. Having taken them before this class, I felt it was a nice summary and refresher course for me, and the extended time spent on the longer courses was very helpful.

Having said that, the Statistical Inference class that I took was its first run, and further installations would very likely improve in subsequent offerings.

Verdict: If you have not taken any stats courses before, perhaps it would be beneficial to spend the extra time on the two Bootcamp courses.