search

gbradburd / conStruct

method for modeling continuous and discrete population genetic structure
35 stars 14 forks source link

about the conStruct restlt #32

Closed guojingfang123 closed 3 years ago

guojingfang123 commented 4 years ago

Hi bradburd, Thank you for your detailed introduction to conStruct. I am trying to run it and I would like to ask two questions about the result.

  1. In the Cross− Validation result, it seems to be stable in K2 and maximum in K3. However, in the layer contribution statistics, the layer contribution is very low. Can I determine whether the optimal K value is 2? cv.pdf spLayerContribution.pdf

2.How do I get the αD value to see if there is isolation by distance?

Thank you, G

gbradburd commented 4 years ago

Hello,

  1. It's possible that you're get strong statistical support in the cross-validation test for extra layers (higher K) that don't contribute much to explaining overall patterns of relatedness. This is especially likely if you have lots of loci. I would interpret your layer contribution plot as support for K=2 or K=3, but there's no hard rule here. Instead, we recommend that users set a threshold of negligible explanatory power (e.g., 5%, or 1%), and reject models with that have layers with contributions below that threshold.

  2. If you look at the documentation for the conStruct function, which you can view using the command help(conStruct), you can see the output of a conStruct call, including where to find either the marginal posterior distribution or maximum a posteriori (MAP) estimate of a specific parameter. But, I would say that you can best assess whether there is isolation by distance by comparing the spatial and nonspatial models for a given value of K. If the spatial model has better support, that is evidence for isolation by distance.

Hope that helps, -Gideon

gbradburd commented 3 years ago

This issue has been inactive for a long while now, so I'm going to close it, but feel free to reopen (or open a new one) if you have further questions.