likelihood VS linkage group length ?

[Etienne Dvrk]

Dear Pasi,

I have a question regarding what to trust more between likelihood and map length.

My data consists of 2 half-sibs families (two crosses with one parent in common) (~2x170 individuals, ~5000 markers).
I already built parental maps with another software and obtained good results, with only a handful of discrepancies with the physical order.

Then I used LepMAP to get a consensus map with all markers together.
First I directly input the assumed physical order like this:

java -cp ./bin OrderMarkers2 evaluateOrder=map_order_physical_LG1_selection.txt data=pcall.txt **improveOrder=0** sexAveraged=1 outputPhasedData=1 minError=0.005 > order_physical_LG1_selection.txt

I get likelihood= -47541 and linkage group length = 153 cM
Then I tried to see if the order could be improved :

java -cp ./bin OrderMarkers2 evaluateOrder=order_physical_LG1_selection.txt data=pcall.txt **improveOrder=1** sexAveraged=1 outputPhasedData=1 minError=0.005 > order_physical_LG1_selection_improved.txt

This time, likelihood= -44670 and linkage group length =184 cM
The likelihood is a bit better but the length is quite higher and there are many small discrepancies with the physical order.

The same phenomenon appears for almost all linkage groups.
So, should I just stick to the physical order, or is the second output potentially more correct?

Thank you,
Etienne D.

PS: my apologies if this was already discussed elsewhere!

[Pasi Rastas]

Dear Etienne,

Thank you for you question.

I think it is normal that order likelihood is lower if you evaluate the map in the physical order compared to denovo order.

If the physical order is correct, using it is easy way to get good maps. Things to check with physical maps:

1) check for outlier individuals based on the number of crossovers (recobinations). Lep-MAP3 ouputs these in the error stream.
2) increase phasingIterations, e.g. to 3. If you notice a change in the likelihood the phasing has improved.
3) if you have uneven distribution of markers (e.g RADseq), proximityScale improves the result.

For truly denovo map, do not include evaluateOrder in OrderMarkers2.

Cheers,
Pasi

[Etienne Dvrk]

Dear Pasi,

Thank you very much for your response and your suggestions.

Increasing phasingIterations has no effect in my case. Anyway the distances computed with the physical order seem fine.

By the way, I could not find an exhaustive documentation of the parameters for each command. For example I had to search in this forum to know if OrderMarkers2 uses Haldane or Kosambi mapping function by default. Does this documentation exist somewhere?

Etienne D.

[Pasi Rastas]

Dear Etienne,

Great that I could be of help.

And sorry, Lep-MAP3 does not have any other documentation than this forum and the program itself. I feel that keeping such a documentation up to date would be too time consuming... And it is very nice that users are not afraid to ask if they are unsure:)

Cheers,
Pasi