CanDIG Variant Search

Francis Nguyen

Application Development Specialist, UHN

May 30, 23

Introduction – Genomic Variants

Genomic variants are differences from one genome to the next. These variants can take multiple forms, from modifications to a single nucleotide, or the deletion, insertion, copying, translocation, inversion, etc., of entire segments. While a genomic variant can technically be relative to anything, in human genetics we generally talk about genomic variants in one of two contexts:

comparison between individuals or groups of individuals to a Genomic Assembly, for example, GRch38.
comparison between the genome of tumour tissue to normal tissue.

These genomic variants are important to study because combinations of them have been implicated in poorer patient prognosis. Genomic risk factors, such as the ones found in 23andme reports, are found by assaying for known variants (usually Single Nucleotide Variants (SNPs)).

Part of what CanDIG wants to enable is making sure genomic data collected by the Marathon of Hope Cancer Centres Network (MoHCCN) is available to researchers.

Data Discovery

CanDIG is a data discovery platform, which means that rather than being interested in just being a store of information, our goal is to enable institutions and consortia to safely share information with authorized researchers. Moreover, CanDIG is deployed at sites, rather than being a centralized store, which allows each local site team to make accessible whatever data they choose for querying and analysis within the federation of CanDIG servers. Due to this federated approach to deployment:

Researchers can still find if there are patients that fit their criteria by asking the federation
Hospitals have complete control over their own databases
Patients can ensure that their data is only used according to the data release policies they have signed off on.

GA4GH’s Beacon protocol propagates queries (such as whether or not there is a specific nucleotide at a specific region on someone’s genome) across a variety of sites that implement Beacon and receive back everybody that matches. CanDIG implements parts of this Beacon protocol in a performant manner.

Data Ingestion

Currently, HTSGet ingests data by creating Data Repository Service objects for each ingested file, which contains metadata about the data for future lookup. We start by considering the entire genome as 10kbp bins, for easier lookup. With the genome being 3.2x10^9 bp in the hg38 assembly, we only have to consider 3.2x10^5 bins, which is a good size for our index. Both these indexes and a mapping for each ingested file’s header to its associated file are stored in a SQLite database.

Suppose we have a region of the genome and two VCFs containing SNVs at those locations in figure 1.

Figure 1: Ingestion diagram. Two different .vcf files containing the SNVs of two different patients are being ingested. Here a small region of the genome is displayed. Each of these features fit inside a 10kbp bin.

Since we use 10kbp bins, we create indexes for each SNV along those bins (see figure 2).

Figure 2: Binning diagram. The ingested features from Figure 1 are indexed, and these indices show which 10kbp bin the feature belongs to. In practice, these features can cross multiple bins, which would add this .vcf to both bins in the index.

This index of bins->VCF, and the headers for each VCF, are now stored in the SQLite database for easier retrieval.

Data Search

When a search comes in, we:

Transform the searched region into buckets, like above
Check which files contain an entry in each of those buckets
Run pysam to grab only the parts of the VCF files that are relevant
Postprocess those files for the Beacon protocol using the headers for each file, also stored in SQLite

Suppose we have a request for a region covered by one of the VCFs above. First, this request is transformed into bins, and we search the index for each file covered by those bins (figure 3).

Figure 3: Search diagram. A search request is transformed into bins, and is intersected with our index. Here, we know that only Patient A's index intersects with the request, so the subsequent `pysam` and other requests can be done.

From the indexes, we know that PatientA.vcf has variants in this region. We can then grab the relevant segment via pysam, and combine it with header information for the metadata.

Frontend

There are two sections to the frontend: a summary landing page (figure 4) and a search page (figure 5).

There are many features to this search page:

A sidebar to filter results, similar to how other genomics sites allow you to filter data, passing these queries to the backend to obtain a paginated list on the frontend.
Data privacy is also respected – if a researcher does not have the credentials necessary to view the patients, counts are returned instead.
As we implement the GA4GH HTSGet protocol, we are interoperable with software that can interface with it (e.g. IGV viewers)

Conclusion

CanDIG allows researchers to find genomic variants of interest in a performant manner, using binning and indexes to speed up searches over the entire genome. These searches are supported by a frontend that shows the data privacy aspect of CanDIG and implements GA4GH HTSGet.

Do you have any questions? Feel free to contact us at [email protected] or on Twitter at @distribgenomics.