Skip to main content

Baselines for Variation


Before we can dive into the types of changes that can occur within a particular individual's genome, we must first answer the question "changes in reference to what?" After all, there is no "average" human's DNA to compare against—genetic code is diverging within individuals and larger geographical populations all the time. How can we compare a single organism's genetic code against a baseline that is more accurately represented as a graph that is constantly evolving?

Reference Genome

At least for the moment, the answer is "we don't". The current solution chooses practicality over the ideal state of things: there is, in fact, a single model known as the reference genome that attempts to recapitulate a consensus representation for the human genome. This consensus model is curated by the Genome Reference Consortium (GRC), and patches to the genome are released on a regular basis to further improve the model as we learn more about the human genome.

Genomes for multiple species are maintained by the GRC, and you can view them on the GRC's website. In the past, the naming conventions were more confusing and versions differed based on where you got the data. However, the academic world has now helpfully synced on the versioning.

For example, the current version of the human genome is GRCh38 where GRC stands for the Genome Reference Consortium, h stands for human, and 38 stands for version 38. The UCSC nomenclature (hg38) is also used interchangeably with this version. Unfortunately, previous versions of the genome versions did not match up on versioning or nomenclature (e.g., hg19 is equivalent to GRCh37 or sometimes GRCh37-lite depending on the context), so you'll need to be careful when using non-current versions of the human genome. If you find yourself using an older version of the reference genome, the Wikipedia page has a helpful table for comparing these historical genome versions.


One of the criticisms of the approach outlined above is the lack of accounting for the diversity amongst the world's populations. For example, ~70% of the material for the Human Genome Project (the foundation for the reference genome) came from an anonymous, white male from Buffalo, NY known only as RP11. Using this linear, haploid reference genome to quantify variation across diverse populations is far from perfect, but (from a scientific perspective) the international genomics community has come up with several strategies for contextualizing genomic variation within this framework (e.g. population frequency databases).

Note that in the next version of the reference genome (hg39), the preferred approach appears to be a graph-based structure that can define variation in a much wider variety of populations. Clearly this is a step forward, but it also introduces significant computational challenges to the processing of data. As such, the correct way to define the next iteration of the reference genome is currently being discussed, and the release is indefinitely on hold per the notice on the Genome Reference Consortium's website.

The Analysis of Tumors vs. Germline Mutations

Having a matched pair of tumor and germline data is especially valuable in variant detection because germline data acts as a precise control. The analyst uses the paring to sort out common but irrelevant single nucleotide variations (SNVs) found in the normal germline tissue from the new cancer driving mutations in the tumor.

Studies that consist only of tumor samples are useful in diagnosing subtypes of the disease. Cancer subtypes often differ in their rate of progression and require distinct treatments. Subtyping helps clinicians provide the precise course of treatment most suited for a given tumor.

Studies that consist only of germline samples are generally useful for discovering genes and variations that are associated with disease. Germline studies are basic research studies where hundreds of thousands of candidate variations are systematically filtered to find a few significant variants. Disease associated variants could be markers for elevated risk or causative mutations that predispose the subject to disease.