Rebooting BCI

[Armand1]

We stopped working on July 19 2019. These are notes on what we need to do to reboot it.

Paper This is the manuscript @ben18785 There is a certain amount of work on the model equations that need to be done. They're not coherent --- see my footnote.

Identify 1) last, best, datasets:

I believe that they are:

2) last, best, code 3) last, best, results:

For the species I believe that these are:

_reproductives_names_stanJuly18.rds _selection_reproductivesJuly18.rds (I probably added the _July18 suffix)

For the functional classes these were (dated July 17): _bci_metaspecies1_freqselection.rds _bci_metaspecies1_indeselection.rds _bci_metaspecies2_freqselection.rds _bci_metaspecies2_indeselection.rds

Where metaspecies1 and 2 designated two different functional classifications.

4) Run (1) on (2) and make sure that they correspond to what we think were the last, best, results (3). And push to Github.

Selection on species

Do we believe the selection coefficients? I believe that we do, but there was some ambiguity on July 19. But there were some minor things that we needed to do:

@ben18785: Could you make all selection estimates, s, relative to the median selection coefficient of the fittest species. Currently they are relative to the upper CI of the fittest species, which has the consequence that all species, even the fittest one, have a negative s.

@ben18785: could you also give me the confidence intervals and estimates of all the betas for each species. I wanted them so that I could calculate absolute fitnesses. They can't be retrieved from the selection coefficients.

@ben18785 can I have the bashed data for any analysis you do.

Selection on functional groups (or clusters) I believe that I have only.... @ben18785: I still need the loos for the result @ben18785: I need the confidence intervals and the estimates for the betas

examining selection on clusters further If these is selection on clusters, then we might look at the fitness of phenotypes directly; estimate a fitness landscape: take the individual species estimates; maybe look at the Price equation. I began doing this.

forward simulations We agreed to do nothing on this at the moment. But Jeronimo might do this in Julia. So we need to talk about that.

[Armand1]

Ben,

There are still several issues with the results.

1) You did not, I think, send me the betas with their CIs. 2) We had a new dataset that you did not use. This was BCI_data_forben_reproductives_withclusters_22July.csv. I should have drawn your attention to this. 3) I also need the bashed data. This is the equivalent of "reproductives_names_stan_July18.rds" I need this to interpret the selection coefficients by actually plotting the data (with their new species designations). 4) I also need the estimates of mu 5) Bear in mind I also need the betas and mu of the clusters

OK -- these things have been sorted thanks @ben18785

[Armand1]

I wanted to check that the selection coefficients make sense in terms of the frequency trajectories. I am concerned here with s_freqinde_50.

The first thing that strikes me is that the s_freqinde_50 does not start at zero and go negative. Rather, there are some strong positive values. So it does not seem to be estimated relative to the median value of the fittest species, but something else --- perhaps the median value of the median species. I think that needs to be fixed.

But setting that aside: does s_freqinde_50 reflect the changes in frequencies? It appears not. I estimated the change in frequencies using a simple linear model, removed those that did not change, and ordered them whether they went up or down in frequency. (From most up to most down, i.e., by the slope). I then colour coded them according to whether s_freqinde_50 was positive or negative. All the positives (blue) should be in the first panels, the reds at the end. But you can see that they're all mixed up. That suggests to me that something is wrong. Or am I misunderstanding something here?

[Armand1]

These issues have now been resolved. There was no fundamental problem, just a difference in accounting between incorporating selection and not. See attached.

checking-estimates-of-selection-on-species_generation-normalised.pdf checking-estimates-of-selection-on-species-against-frequencies.pdf