8

We have whole genome sequencing data for patients (not-cancer) (n=60) and for healthy controls (n=20). The sequencing centre has provided us with the best practice bioinformatics analyses including reads mapping (.BAM) and variant calling using GATK (.vcf) as well as annotation (annotated .vcf and .gVCF).

What should be our next steps? We are interested to see if there are any differences (global and/or specific) between the groups.

jessica
  • 81
  • 2
  • 1
    Is that really what you're interested in, or would you rather find likely causative differences? – Devon Ryan Jun 05 '17 at 18:11
  • 1
    Hi Jessica, thanks for writing a question up for Bioinformatics Stack Exchange. Your question fits well with the format we like on this site, in that it is specific to a particular task, and includes a bit of a story behind it.
    If you would like to improve the quality of answers for this question, my suggestion would be to add in information about what types of analyses you are already familiar or comfortable with. Have you used GATK or R before? Is this a diagnostic setting, where a structured, documented, and replicable analysis is important?     – gringer Jun 05 '17 at 19:36
  • Sounds like you would like to do GWAS, but for that you would probably need way more samples. Can you subset your variants to test only relevant regions of genome? That would help you to get a statistical power (less tests -> better). – Kamil S Jaron Jun 05 '17 at 20:57

2 Answers2

2

I'm not familiar with the program, but apparently Hail is setting itself up as a swiss-army chainsaw project for doing downstream analysis on variant-called datasets.

An overview of Hail can be found here:

http://blog.cloudera.com/blog/2017/05/hail-scalable-genomics-analysis-with-spark/

A tutorial on association testing can be found here:

https://hail.is/hail/tutorial.html#Association-testing

gringer
  • 14,012
  • 5
  • 23
  • 79
2

If you have gVCFs, the first thing you should try is joint variant calling. According to GATK, joint variant calling "empowers variant discovery by providing the ability to leverage population-wide information from a cohort of multiple sample[sic], allowing us to detect variants with great sensitivity and genotype samples as accurately as possible." source.

Once you have two sets of high-quality variants, what you do next depends on your research question. Are you looking for druggable mutations? Underlying causes? Biomarkers? Patient prognosis? You can look at the most frequently mutated positions in the patient cohort and compare them to your reference cohort, check for co-occurring mutations, cluster them, do principal component analysis, do some machine learning to stratify them, etc.

morgantaschuk
  • 530
  • 4
  • 9