Should a Rich Genome Variant File End the Storage of Raw Read Data?

Deep sequencing of a human genome produces large datasets sized in the hundreds of gigabytes, while the associated genome variant file measures about half a gigabyte. Now that we are sequencing thousands of human genomes has the time come to store and use only this variant file?

Generation of an accurate and comprehensive genome variant file is a critical first step in extracting biological knowledge from sequence data. This file lists all the sequence differences found in the analyzed genome contrasted to a commonly-used human reference genome. Researchers compare files from a set of genomes to find the genetic variations that might lead to a specific disease. Since the early days of Complete Genomics’ human genome sequencing service, we have provided these research-ready genomic reports. Scientists have thereby obtained the data from complete human genome sequencing without the overhead and expense of large computational facilities to handle, map, and assemble raw reads—even though the raw data was always supplied as well.

In fact, the first peer-reviewed papers published using our service employed the variant files and didn’t require the raw read data. The Institute for Systems Biology (ISB) sequenced a family where both children suffered from skeletal malformations caused by Miller syndrome, while both parents were unaffected. By sequencing these four genomes and comparing the variant files, ISB was able to identify the causal variant.¹ The University of Texas Southwestern Medical Center used Complete Genomics’ service to solve a clinical case in which an 11‑month-old child presented with a very high cholesterol level (LDL‑C of 837). A standard blood test reported the infant was not lacking the protein that would bring down the cholesterol level. However, the variant file data revealed that, in fact, the protein must be absent.²

Preserving raw data is always a good policy when using a new technology. Initially, the sequencing cost per genome was high, and studies rarely included more than a dozen genomes. Structural variants and copy number variants were not reported and the accuracy of the reported variants was either unknown or not trusted by the research community. Now, after two years of intense R&D and a substantial reduction in sequencing price, what were good reasons in the past may be unnecessary limitations today. Is it wise now to store only the variant file and delete all the raw data? Yes—but only if the variant file is rich enough and accurate enough.

At Complete Genomics, by using more than 40x read coverage and local-de novo assembly, we provide a rich variant file that describes and fully annotates SNPs, indels, substitutions, copy number variants, structural variants, and mobile element insertions for virtually the entire genome. In addition, every reported variant has a precise confidence score which researchers can use to obtain the desired tradeoff between sensitivity and specificity. The increased comprehensiveness of variant detection, and flexibility to adjust the balance between false positives and false negatives offered by the variant files, mitigate much of the need to store an enormous amount of raw read data including a quality score for each base in each read.

I have heard the argument that it is important to save raw read data for reanalysis in the future with improved software. But wouldn’t it be cheaper to re-sequence and re-annotate selected genomes rather than to pay for storing raw read data for all genomes in all studies? Another argument offered in favor of keeping the reads is that scientific journals might require them as a condition of publication. This requirement is based on yesterday’s technology. Journals should instead require a rich variant file as support for biological and medical conclusions. The research community should lead a discussion about format and content standards for such a data file. In the end, there remains only one case where I would keep raw data—when the DNA samples initially sequenced are so small that re-sequencing is not an option.

The concern of many researchers dealing with the complexities of the large datasets represented by the raw read data file could practically vanish; they can work with the smaller and simpler-to-use variant file complemented by a rapidly improving set of software tools. This will greatly simplify and reduce the cost of larger studies which will involve hundreds or thousands of complete genomes. These large studies will be required to fully identify and understand molecular pathologies leading to human diseases.

Advances in the completeness, accuracy, and confidence levels of our genome variant files, coupled with the affordable sequencing price, make it acceptable to save and use only the rich and simple-to-use, research-ready variant file.

^{1 Roach JC, Glusman G, Smit AFA et al.: Analysis of genetic inheritance in a family quartet by whole-genome sequencing. Science 328(5978), 636-639 (2010).}

^{2 Rios J, Stein E, Shendure J, Hobb HH, Cohen JJ: Identification by whole-genome resequencing of gene defect responsible for severe hypercholesterolemina. Hum. Mol. Genet. 19(22), 4313-4318 (2010).}