High similarity sample clustering

[dkoslicki]

Get all pairs of samples with Jaccard =1, >= 0.99, and >=0.95. Get PW matrix from this and cluster it. See what it looks like (bunch of small distant clusters? Larger groups? What does coloring by metadata look like? etc.)

[dkoslicki]

Next todo:

1. Get a list of all the distinct accessions that have a Jaccard of 1 with another accession
2. Compute the all pairwise Jaccards between these accessions
3. See if there are any values besides 0 and 1 in it (i.e. test the question: "Are these samples at least partially similar to other samples with this feature, or are they all distinct from each other?")
4. Compute the matrix of (all accessions) x (accessions that have a Jaccard of 1 with another accession matrix)
5. See if that matrix has values other than 0 or 1 ("are these similar to anything else?")
6. If the answer to step 5 is yes, then cluster them.

Additionally:

1. Add metadata information (bioproject, sequencing center, date, geo location, etc.) and see if each Jaccard-of-1 cluster has consistent or inconsistent metadata.