Dear Hyphys,

I have some questions regarding the KH test applied after a GARD run (RecombinationProcessor.bf). Let's do it with some examples:

Gene #1

5 partitions were found by GARD:
- Partition 1: 0-308
- Partition 2: 309-426
- Partition 3: 427-813
- Partition 4: 814-998
- Partition 5: 999-1922

From the RecombinationProcessor.bf output we obtain:


Code (]
[...):

KH Testing partition 1 Tree 2 base LRT = 12.8929. p-value = 0.0758 [...] KH Testing partition 2 Tree 1 base LRT = 49.6799. p-value = 0.0001 Tree 3 base LRT = 8.68062. p-value = 0.0001 [...] KH Testing partition 3 Tree 2 base LRT = 59.6993. p-value = 0.0001 Tree 4 base LRT = 24.3077. p-value = 0.0001 [...] KH Testing partition 4 Tree 3 base LRT = 16.3848. p-value = 0.0001 Tree 5 base LRT = 62.1624. p-value = 0.0001 [...] KH Testing partition 5 Tree 4 base LRT = 10.3781. p-value = 0.0001 A total of 3/4 significant couplings Mean splits identify: 0.165

Gene #2

2 partitions:
- 0-807
- 808-1013

RecombinationProcessor.bf output:


Code (]
[...):

KH Testing partition 1 Tree 2 base LRT = 23.8159. p-value = 0.1265 [...] KH Testing partition 2 Tree 1 base LRT = 104.634. p-value = 0.0001 A total of 0/1 significant couplings Mean splits identify: 0.25

So, with the first gene, all the test are significant (at the 0.05 level), except when we fit the topology of the partition 2 on the partition 1. Basically, partition 1 has not a significantly different history than partition 2, but partition 2 has one.  ???

The same is observed with the gene #2.

In those case what should we do? Considered that if 2 is significantly different than 1, that implies that 1 is different than 2, and that the test was not significant because of a lack of power (e.g. small partition)?

By the way, what does the "A total of x/x significant couplings" and "Mean splits identify: xxx" mean?

Thank you,
-Tristan

Dear Tristan,

"A total of x/y significant couplings" means that among 'x' breakpoints, 'y' had significant KH tests on BOTH sides of the breakpoint. "Mean splits identify: xxx" means that if you consider all N+1 trees (for N breakpoints), the splits (bi-partitions of taxa induced by each internal node) are 'xxx%' the same between trees on average. This gives you a rough idea of how many sequences are involved in recombination events (small number - more sequences, high number - more sequences).

In terms of one of the comparisons being significant, the most likely explanation is that one of the two partitions has really different branch lengths, but not that different of a topology. Short partitions (like your partition 2 in Gene 1) tend to collapse a lot of branches to 0, i.e. there may be an unusual stretch of conserved nucleotides in the region, which does not necessarily imply a different topology.

Cheers,
Sergei

Thank you Sergei,
So just to be sure I follow you, for gene #1 you only consider 4 partitions (1+2, 3, 4, 5), and for gene #2, you conclude that there is no evidence of recombination, correct?


-Tristan

Dear Tristan,

Recombination is not necessarily accompanied by the change in the tree topology. For example, distorted branch lengths could also result from recombination. However, if one wishes to be as conservative as possible, and demand that strong evidence of discordant phylogenetic signal be present among partitions, then indeed, gene 1 would have 4 fragments, and gene 2 would have 1 (i.e. no recombination).

Cheers,
Sergei