Individual chapters of the Next Generation DNA Sequencing book are now on sale as downloads from the Cold Spring Harbor Press website:

http://cshlpress.org/ngs/

I am really quite enthusiastic about this. Very few (if any) other books sell individual chapters as downloads. Hopefully some people will be willing to spring for $7 for an interesting or relevant chapter and then decide they want to get the whole  book.

Genotyping on MiSeq with overlapping paired-end reads

Our Illumina MiSeq can now do 2 X 250 bp paired-end reads. We are going to use an amplicon method for a simple genotyping assay to identify very rare mutations at a single SNP site. The idea is to amplify a short fragment (less than 300-400 bp) so the F and Rev reads overlap by 50 or 100 bp, not just by a few bp, as we have done in some other assays (such as 16S metagenomics), and put our target SNP in the center of the overlap region. In this way, the SNP will be located in a high quality region of both reads, and we can get the maximum accuracy in the genotype call.

One problem with using the MiSeq for amplicon assays is that it is very sensitive to the base-pair composition at each cycle. Illumina tech support insists that a spike-in of 50% Phix DNA is necessary for any amplicon assay. (Our Genomics lab tried it with 40% added Phix and got poor results.)  However, with the latest upgrades, the MiSeq is now producing 15 Million reads per run, so with the 50% Phix spike-in, it still produces over 7 M usable (high-quality) reads.


In the past, we have not used Illumina to test for rare mutations (ultra-deep sequencing) because the overall error rate is in the 0.3-0.5 range. Even if we restrict results to bases >Q30 at our target, we can't find mutations with a certainty below the error frequency of about 1 per thousand - since that would be the expected rate of false positives. With overlapping high quality reads, we can recalculate a new Q-score based on both reads (assuming that they both agree on a variant call). I am still thinking about the proper math for this (a joint probability of error), but it is something similar to the sum of the two Q-scores (product of the two error frequencies).  This would allow us to find mutations in the one per ten thousand  (or even hundred thousand) range with a low false positive rate.  The PANDA-seq program uses similar math to calculate Q-scores for overlapping regions, so we are going to use that for the first pass on this data.

Martin Blaser speaks about the early-life Microbiome

I like working at NYU. Interacting with very smart people has its own benefits. Even if their smarts don't directly rub off, at least I can learn something from listening to their ideas. Martin Blaser is a remarkably accomplished scientist (Past President of the Infectious Disease Society, member of the Institute of Medicine of the National Academy of Sciences, chair of the NIH Advisory Board for Clinical Research, and author of about 500 publications) with wide ranging interests in infectious disease. He has made substantial contributions to the study of Helicobacter pylori and its role in human diseases including ulcer and gastric cancer, as well as esophageal cancer, Crohn's disease and inflammatory bowel disease. Dr. Blaser has been an important contributor to the Human Microbiome Project. He has been recently profiles in the New York Times and the New Yorker magazine. Today he gave the Dean's Lecture at NYU Medical Center, where he discussed recent research in his lab on the effects of antibiotics on the microbiome of mice and the implications for human infants and children. 

Low levels of antibiotics are commonly used to supplement the feed of livestock animals, causing them to grow faster, more efficiently convert food to body mass, and to have higher levels of body fat. In a mouse system, the Blaser Lab has shown that steady low levels of antibiotics from birth to 28 weeks of age produce substantial changes in the microbe composition of the gut, as well as an increase in body fat.  The antibiotic altered microbiome, when combined with a high fat diet, has a synergistic effect on body fat (higher than either treatment alone). A similar effect is observed when several short pulses of antibiotic are administered at doses similar to what is used to treat infections in human children. The altered microbiome in antibiotic treated mice was also associated with immune suppression in the gut (lower levels of T-cells, lower levels of immune system molecules such as interleukins). 

Connecting back to human epidemiology, Americans have become much more obese in the past 50 years. In 1962, about 14% of Americans were obese, but by 2010, 42% of Americans have become obese. Blaser's point is that diet and exercise have not changed dramatically in the past generation (we ate almost as badly and were almost as lazy back in the 60's and 70's), but antibiotic use has changed dramatically.  Typical children now receive 10-20 courses of antibiotic treatments. Coincidentally, the prevalence of H. pylori in the gut has decreased  from over 50% down to just 5% of Americans.  The immune system changes that were observed in the antibiotic treated mice may correlate with the huge increases that are now observed in human asthma, celiac disease inflammatory bowel disease, type 1 diabetes, and other forms of allergy and autoimmune disorders.  Of course correlation does not prove causation, but the work of Blaser and others is beginning to identify  mechanisms that connect changes in the microbiome to these disorders. In mice, infection with H. .pylori has been shown to protect from respiratory allergic reactions.

Thanks to modern medicine, Americans now live longer, are much less likely to die of routine infections, and are taller, but we are much more obese, have more cardiac disease, and more autoimmune disease.  So, can we create a different equilibrium with our microbiome that retains the life saving anti-infection benefits of modern antibiotics, but restores the metabolic and immune balance from an earlier era?

Australian NGS Tutorial

These Galaxy-based NGS tutorials from the Australian "Super Science"  Genomics Virtual Lab are really excellent:

https://genome.edu.au/wiki/Galaxy_Tutorials

Very clear explanations, nice screen shots, and a hands-on demonstration for students to see for themselves why biological replicates are necessary to find differential expression with RNA-seq.

Its hard to make NGS simple

Our Informatics group has been running training seminars for some postdocs and upper level grad students to teach computer skills for NGS data analysis. This was at their request, not my idea. What the students really wanted was the ability to run the common NGS software on their own, rather than be completely dependent on the bioinformatics group to do all of the work for them.  Our HPC director, Efstratios Efstathiadis, and I made up some lectures on basic Unix command-line skills and tried to make some simple workflows to demonstrate common NGS applications such as alignment, visualization, ChIP-seq, and RNA-seq.

The first tutorial went pretty well. We created an Amazon Cloud virtual machine and make accounts for all of our students (it had to be on the Cloud since we still have no local servers - see hurricane story in my previous blog post). We installed BWA, SAMtools, and the necessary supporting modules. We uploaded a very small FASTQ file as a practice data set and a reference genome. Then in class, we taught about a dozen basic Unix commands for the complete novice (pwd, ls, cd, mkdir, man, rm, cp, mv, head, more, ...). Then we  had the students align the sample data with BWA, transform to sorted indexed BAM files with SAMtools, then download (.BAM and .BAI files) and visualize the final data on their own computers with IGV. This took about 2.5 hours for a class of 10 students, and was generally felt to be a solid success by all participants.

The second tutorial was supposed to be a bit closer to a real use case for a ChIP-seq experiment. We created a somewhat larger practice data set that contained data just from human chromosome 1, with 2 FASTQ files: a ChIP sample and an input control (about 1 million reads per file). I gave a little lecture on the basics of ChIP-seq technology, how peak calling software works, and what to look out for in terms of QC of the data. Then we had the students make the alignments, and transform into BAM format plus index for the two FASTQ files. Then run MACS to find peaks and produce a BED file as output. Then download BAM, BAI, and BED files to local laptops and visualize on IGV as well as load BED file into Galaxy and compare with locations of known gene start sites.

This tutorial turned out to be overly ambitious. The amount of CPU churning required for 10 students to run 2 BWA jobs each on 1 million read FASTQ files was much more than our standard Amazon EC2 VM could support. We should have created all of the output files in advance and shared them with the students. Then they could start each job, kill the job, and move on. Also, downloading all the final data from the EC2 instance was a hassle. We should have just passed around a USB drive with the final BAM files. This class went way beyond our 2.5 hours, so we never found the time to show the students how to load all gene Transcription Start Sites into Galaxy and overlap with the BED file to annotate ChIP-seq peaks with respect to promoters (one of my favorite tricks).

I'm supposed to follow up with an RNA-seq tutorial in a couple of weeks, and I doubt that I can make TopHat/Cufflinks simple enough to run smoothly in a classroom setting. Overall, I have learned that it is darn hard to make even routine NGS tasks simple and bullet proof. I am leaning more toward some type of Galaxy on the Cloud solution for lab scientists who want to take some control over their data analysis tasks.