Gene trait matching

Gene-trait matching is an effective approach to identify genes that could be responsible for an observed phenotype.

The principle is simple: If one group of genomes has a specific trait that another does not, search for orthologous genes (or functional annotations) that occur in one group and are absent in the other.

Usage

Open the gene trait matching page and enter two non-intersecting groups of genomes, for example: Lactobacillaceae vs Propionibacteriaceae and Streptococcaceae .

Gene trait matching runs as a background job. After submitting, you can close the tab or navigate elsewhere; the calculation continues on the server. Results are saved and can be reopened from the Job Queues page at any time.

The resulting table can be downloaded in CSV format through the settings sidebar.

Example use case

The following example is based on a real experiment with 39 strains from the same microbial genus. 23 strains can metabolise a specific compound (green), the others (red) can not.

To find the responsible gene(s), open the gene-trait-matching view, define the two groups of genomes, and click on 'Submit'.

In this case, gene-trait matching found a strong correlation of the trait with a small number of orthologs. A closer look indicated that these orthologous genes were always located close to each other on the genome. A follow-up RNA-Seq experiment also showed a link between the phenotype and this gene cluster.

This experiment was ideal for gene-trait matching because the strains were relatively closely related and the trait was distributed amongst different clusters. If this is not the case, gene-trait matching is less likely to work. For example, had the phenotype been strongly correlated with the phylogenetic clusters, like in the image below, too many genes would probably have shown up as significantly different between the two groups.

Background

By default, Arx will run a Fisher's exact test for each orthologous gene and apply Benjamini/Hochberg multiple testing correction (alpha = 10 %).

Caution: It is not guaranteed that the assumption of Fisher's test, that all isolates have a random and independently distributed probability for exhibiting each state, is valid because of population structure. See the example above.

Advanced usage

In the settings sidebar (wheel on top right), it is possible to change:

• Test method
  • fast-fisher: Faster, custom implementation of Fisher's exact test. Original algorithm by painyeph, Cython implementation by me (Thomas Roder)
  • fisher: Reference implementation in the scipy library, should give almost exactly the same result as fast-fisher, but about 100x slower than fast-fisher
  • boschloo: Boschloo's exact test is a more powerful variant of Fisher's exact test that can be used when the column sums of the contingency table are known in advance.
  • scoary: Scoary-2 with phylogenetic correction (see below)
• The category of annotations to use
• The multiple testing algorithm and associated alpha value

Scoary-2: phylogenetic correction

When the scoary method is selected, Arx uses Scoary-2 to account for population structure. Standard Fisher's test assumes all isolates are independent, but closely related strains share evolutionary history, so a gene that is merely co-inherited with the trait in a single clade can appear significant even if it has no causal role.

Scoary-2 corrects for this by mapping genes and traits onto the phylogenetic tree and computing an empirical p-value based on how often the observed association pattern could arise by chance under the tree topology. This makes the results more conservative and biologically meaningful when working with phylogenetically structured datasets.

Interactive results viewer

Scoary-2 results open in an interactive viewer. For each significant gene, you can explore:

• The per-genome presence/absence heatmap
• The phylogenetic tree with trait and gene overlaid
• Per-gene statistics and annotation details

For more information, read the paper Brynildsrud et al, Genome Biol, 2016 about the original Scoary tool.