run_DDD

[bheimbu]

Dear Fabien,

I'm using your run_DDD script and encounter following error:

[1] "Linear dependence of extinction rate" You are optimizing lambda mu K You are fixing nothing Optimizing the likelihood - this may take a while. The loglikelihood for the initial parameter values is -207.7032 Error: in ‘subplex’: The requested tolerance (tol=1e-07) is too small for mxrej=10. Called from: value[[3L]](cond)

Have you ever seen this before? My tree has 79 tips, is ultrametric and the known clade size is 114. I used various values for total_richness (150, 200, 250) but always get the same error.

Any help is highly appreciated.

Cheers Bastian

[FabienCondamine]

Dear Bastian,

Thanks for your interest in this code and asking your question.

This part of the code calls the R package DDD, developed by Rampal Etienne among others, and I also got this error message for some phylogenies recently. Sometimes it can be fixed with different starting values for the parameters, sometimes not.

I can't really know because I am not the developer of DDD. There are other options in the function, which have not been modified in my code. 'subplex' is an optimization algorithm and perhaps you can change this. My code calls for the function of the package but does not change it. Look in the pdf of DDD package https://cran.r-project.org/web/packages/DDD/DDD.pdf. Otherwise, if you don't find anything, perhaps you can ask Rampal directly.

Sorry for not being very helpful. I hope you will find a way. Best wishes,

Fabien

Fabien L. Condamine, Ph.D.

CNRS http://www.cnrs.fr/index.php/en, UMR 5554 Institut des Sciences de l'Evolution de Montpellier http://www.isem.univ-montp2.fr/en/ Team Phylogeny and Molecular Evolution https://scholar.google.fr/citations?user=i_bc4KQAAAAJ&hl=en *Université de Montpellier, *Bât. 22 RDC, CC 064 Place Eugène Bataillon 34095 Montpellier Cedex 5 France

Personal website: www.fabiencondamine.org http://www.fabiencondamine.org

Le mar. 10 nov. 2020 à 09:12, bastianheimburger notifications@github.com a écrit :

Dear Fabien,

I'm using your run_DDD script and encounter following error:

[1] "Linear dependence of extinction rate" You are optimizing lambda mu K You are fixing nothing Optimizing the likelihood - this may take a while. The loglikelihood for the initial parameter values is -207.7032 Error: in ‘subplex’: The requested tolerance (tol=1e-07) is too small for mxrej=10. Called from: value[3L]

Have you ever seen this before? My tree has 79 tips, is ultrametric and the known clade size is 114. I used various values for total_richness (150, 200, 250) but always get the same error.

Any help is highly appreciated.

Cheers Bastian

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/FabienCondamine/Diversification_analyses/issues/1, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABT5APHYFEQXFVJEJS6N2W3SPDYWHANCNFSM4TQKWQDA .

[bheimbu]

Dear Fabien,

thanks for your reply. I will check out their manual first. Fingers crossed that I will something to fix this.

Cheers Bastian

[bheimbu]

Dear Fabien,

I have changed the optimization method from subplex to simplex and it worked perfectly. You may give it a try.

I have a few other questions, if you don't mind:

How can I include a constant-rate (CR) birth–death model and/or a Yule model in DDD? I have found two papers, where they use them in DDD, but I don't know how they did it. I have actually asked them personally but received no reply yet.

And another question concerning my results: Is it normal to see -1 values for Lambda, Mu etc?

Cheers Bastian

[FabienCondamine]

Dear Bastian,

Good you tried to change the method. I did try several times, and in other functions.

But here what you tell me is not a good sign. -1 values mean the model did not work. So perhaps try to change the initial values of the parameter now.

Good luck. Best regards,

Fabien

Fabien L. Condamine, Ph.D.

CNRS http://www.cnrs.fr/index.php/en, UMR 5554 Institut des Sciences de l'Evolution de Montpellier http://www.isem.univ-montp2.fr/en/ Team Phylogeny and Molecular Evolution https://scholar.google.fr/citations?user=i_bc4KQAAAAJ&hl=en *Université de Montpellier, *Bât. 22 RDC, CC 064 Place Eugène Bataillon 34095 Montpellier Cedex 5 France

Personal website: www.fabiencondamine.org http://www.fabiencondamine.org

Le mer. 11 nov. 2020 à 09:23, bastianheimburger notifications@github.com a écrit :

Dear Fabien,

I have changed the optimization method from subplex to simplex and it worked perfectly. You may give it a try.

Just a quick question concerning my results https://github.com/FabienCondamine/Diversification_analyses/files/5522678/results_DDD.txt: Is it normal to see -1 values for Lambda, Mu etc?

Cheers Bastian

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/FabienCondamine/Diversification_analyses/issues/1#issuecomment-725279211, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABT5APDNZMVNPDMENRCGBITSPJCYNANCNFSM4TQKWQDA .

[bheimbu]

Hi Fabien,

thanks for your reply. Can you make an educated guess about the initial values or is it more a thing of 'trial & error'?

Cheers Bastian

ps: Do you know how to implement the Yule and CR models in DDD? Just asked this also in my last comment ;)

[FabienCondamine]

Hey,

Honestly, values like 0.1 for lambda and 0.01 for mu should work a broad range of clades, but who knows. So yes, there is some trial, fail, and try again :).

I don't fit the Yule or constant BD with the DDD package, but rather with RPANDA, which are actually included in the run_Morlon_models() function.

I hope this helps.

Cheers,

Fabien

Fabien L. Condamine, Ph.D.

CNRS http://www.cnrs.fr/index.php/en, UMR 5554 Institut des Sciences de l'Evolution de Montpellier http://www.isem.univ-montp2.fr/en/ Team Phylogeny and Molecular Evolution https://scholar.google.fr/citations?user=i_bc4KQAAAAJ&hl=en *Université de Montpellier, *Bât. 22 RDC, CC 064 Place Eugène Bataillon 34095 Montpellier Cedex 5 France

Personal website: www.fabiencondamine.org http://www.fabiencondamine.org

Le mer. 11 nov. 2020 à 14:15, bastianheimburger notifications@github.com a écrit :

Hi Fabien,

can you make an educated guess about the initial values or is it more a thing of 'trial & error'?

Cheers Bastian

ps: Do you know how to implement the Yule and CR models in DDD? Just asked this also in my last comment ;)

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/FabienCondamine/Diversification_analyses/issues/1#issuecomment-725416497, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABT5APBB3CV2EFHOFEMU4JLSPKE5NANCNFSM4TQKWQDA .

[bheimbu]

Many thanks again,

this chat with you is invaluable to me. I will try your suggestions and let you know asap.

Cheers Bastian

[bheimbu]

Hi Fabienne,

your suggestions did the trick. I will definitely reference your pipeline in my paper.

Many thanks,

Bastian

[FabienCondamine]

Hi Bastian,

Thanks a lot! I'm glad it helped and that you solved the issue.

Have a good afternoon,

Fabien

Fabien L. Condamine, Ph.D.

CNRS http://www.cnrs.fr/index.php/en, UMR 5554 Institut des Sciences de l'Evolution de Montpellier http://www.isem.univ-montp2.fr/en/ Team Phylogeny and Molecular Evolution https://scholar.google.fr/citations?user=i_bc4KQAAAAJ&hl=en *Université de Montpellier, *Bât. 22 RDC, CC 064 Place Eugène Bataillon 34095 Montpellier Cedex 5 France

Personal website: www.fabiencondamine.org http://www.fabiencondamine.org

Le jeu. 12 nov. 2020 à 13:37, bastianheimburger notifications@github.com a écrit :

Hi Fabienne,

your suggestions did the trick. I will definitely reference your pipeline in my paper.

Many thanks,

Bastian

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/FabienCondamine/Diversification_analyses/issues/1#issuecomment-726052313, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABT5APEKVVXFRPMIXIMQBVDSPPJKLANCNFSM4TQKWQDA .

[bheimbu]

Hi Fabien,

I realize more and more that your research interests are overlapping with mine. If you don't mind, I would like to hear your thoughts about a recent finding of mine:

I did a lineage-through-time plot of a termite group in Australia and employed a MCCR test (γ statistic = -2.496, p = 0.0001, 10k simulations) and model-fitting (see above) using various sampling fractions and in all cases diversity-dependent diversification is indicated. However, BAMM analysis shows no signs of rate shifts associated with the two pulses observed in the LTT plot. May it be that my tree is just too small including only 79 taxa!? In other words, those results are not contradicting themselves, right?

Cheers Bastian

[FabienCondamine]

Hey,

Thanks for sharing this. The gamma statistics could also be biased, as the other models. So it's important to perform multiple models to cross-validate the results. That's a pity BAMM does not recover the shifts, but does it recover the slowdown in diversification? If yes, then you are good to go: all the models indicate decrease in diversification through time, and DDD tells this diversification is likely linked to clade's diversity.

Cheers,

Fabien

Fabien L. Condamine, Ph.D.

CNRS http://www.cnrs.fr/index.php/en, UMR 5554 Institut des Sciences de l'Evolution de Montpellier http://www.isem.univ-montp2.fr/en/ Team Phylogeny and Molecular Evolution https://scholar.google.fr/citations?user=i_bc4KQAAAAJ&hl=en *Université de Montpellier, *Bât. 22 RDC, CC 064 Place Eugène Bataillon 34095 Montpellier Cedex 5 France

Personal website: www.fabiencondamine.org http://www.fabiencondamine.org

Le jeu. 12 nov. 2020 à 15:33, bastianheimburger notifications@github.com a écrit :

Hi Fabien,

I realize more and more that your research interests are overlapping with mine. If you don't mind, I would like to hear your thoughts about a recent finding of mine:

I did a lineage-through-time plot https://github.com/FabienCondamine/Diversification_analyses/files/5531216/LTT.pdf of a termite group in Australia and employed a MCCR test (γ statistic = -2.496, p = 0.0001, 10k simulations) and model-fitting (see above) using various sampling fractions and in all cases diversity-dependent diversification is indicated. However, BAMM analysis shows no signs of rate shifts associated with the two pulses observed in the LTT plot. May it be that my tree is just too small including only 79 taxa!? In other words, those results are not contradicting themselves, right?

Cheers Bastian

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/FabienCondamine/Diversification_analyses/issues/1#issuecomment-726114683, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABT5APBTVY5UZ3YDUD5ZRHDSPPW3VANCNFSM4TQKWQDA .

[bheimbu]

Hi,

there is a marked slowdown in terms of speciation rates. The plot shows three curves, because there are diverging clades within this termite group.

Though following your comment, this would mean I am good to go?

Cheers Bastian

[FabienCondamine]

Hey,

Yes, that's good because the gamma stat indicates a slowdown, DDD tells there is decline in speciation due to the clade's diversity, and BAMM tells that speciation is decreasing through time as well. So: all the same ;-)

Cheers,

Fabien

Fabien L. Condamine, Ph.D.

CNRS http://www.cnrs.fr/index.php/en, UMR 5554 Institut des Sciences de l'Evolution de Montpellier http://www.isem.univ-montp2.fr/en/ Team Phylogeny and Molecular Evolution https://scholar.google.fr/citations?user=i_bc4KQAAAAJ&hl=en *Université de Montpellier, *Bât. 22 RDC, CC 064 Place Eugène Bataillon 34095 Montpellier Cedex 5 France

Personal website: www.fabiencondamine.org http://www.fabiencondamine.org

Le jeu. 12 nov. 2020 à 16:17, bastianheimburger notifications@github.com a écrit :

Hi,

there is a marked slowdown in terms of speciation rates https://github.com/FabienCondamine/Diversification_analyses/files/5531470/speciation_rate.pdf. The plot shows three curves, because there are diverging clades within this termite group.

Though following your comment, this would mean I am good to go?

Cheers Bastian

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/FabienCondamine/Diversification_analyses/issues/1#issuecomment-726142204, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABT5APEILKO66IVIZ3SCJZDSPP4BBANCNFSM4TQKWQDA .

[bheimbu]

Many thanks,

for your comments. I am still wondering why BAMM shows no rate shifts, but greater sampling effort may change this in the future.

Have a nice evening and cheers Bastian

[bheimbu]

Hi Fabien,

I have done a couple of more runs using various initial values and realized that for total_richness=250 the carrying capacity drops to 3.58 and the AICc value is skyrocketing. But I don't know why? Here you can see the results of three runs using initial values of 0.1 and 0.01 for lambda and mu, repsectively.

Might it be a problem again with the optimmethod?

Cheers Bastian

[FabienCondamine]

Hi Bastian,

Thanks for letting me know.

The K value for the DDX+E model is indeed very strange. Because the clade is much richer than this K, it is unlikely such a model would fit well the data. This seems to be the case given the large difference in AIC between the two models (∆AIC=86). Hence, the model with K=3.58 should be discarded in favor of the model with K=352.

Have you tried this model with other values of lambda and mu? Perhaps the K will change accordingly.

And also, why are you changing the total_richness? This must be set to the total number of species in your clade. Have you an uncertainty about the species number in your clade?

All the best,

Fabien

Fabien L. Condamine, Ph.D.

CNRS http://www.cnrs.fr/index.php/en, UMR 5554 Institut des Sciences de l'Evolution de Montpellier http://www.isem.univ-montp2.fr/en/ Team Phylogeny and Molecular Evolution https://scholar.google.fr/citations?user=i_bc4KQAAAAJ&hl=en *Université de Montpellier, *Bât. 22 RDC, CC 064 Place Eugène Bataillon 34095 Montpellier Cedex 5 France

Personal website: www.fabiencondamine.org http://www.fabiencondamine.org

Le mar. 1 déc. 2020 à 16:03, bastianheimburger notifications@github.com a écrit :

Hi Fabien,

I have done a couple of more runs using various initial values and realized that for total_richness=250 the carrying capacity drops to 3.58 and the AICc value is skyrocketing. But I don't know why? Here https://github.com/FabienCondamine/Diversification_analyses/files/5623391/final_ddd.txt you can see the results of three runs using initial values of 0.1 and 0.01 for lambda and mu, repsectively.

Might it be a problem again with the optimmethod?

Cheers Bastian

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/FabienCondamine/Diversification_analyses/issues/1#issuecomment-736610104, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABT5APFIURHVFRCVMGXX6HDSSUAVVANCNFSM4TQKWQDA .

[bheimbu]

Hi Fabien,

Have you tried this model with other values of lambda and mu? Perhaps the K will change accordingly.

Indeed, I tried also the default values. Here are the results. But as you can see, for total_richness=250 the K and AICc values are as strange as above. Tomorrow I will try other initial values (lambda = 0.01 and mu = 0.001), and maybe I am trying another value for total_richness.

And also, why are you changing the total_richness? This must be set to the total number of species in your clade. Have you an uncertainty about the species number in your clade?

Yes, it is totally unknown how many species there are. That's why, I'm testing three different assumptions in regards to the proportion of missing species within the clade/phylogenetic tree: I have used 114 species (the presumed number of species, but definitely way too low), 150 species (again, too low), and 250 species (more likely but maybe too high). And as such it seems that the DDX model seems to fit best compared over all assumptions. 150 and 250 species are based on preliminary species delimitation results.

Cheers Bastian

[FabienCondamine]

Hello Bastian,

Please keep me posted for your new tests.

We should find a solution ;-)

All the best,

Fabien

Fabien L. Condamine, Ph.D.

CNRS http://www.cnrs.fr/index.php/en, UMR 5554 Institut des Sciences de l'Evolution de Montpellier http://www.isem.univ-montp2.fr/en/ Team Phylogeny and Molecular Evolution https://scholar.google.fr/citations?user=i_bc4KQAAAAJ&hl=en *Université de Montpellier, *Bât. 22 RDC, CC 064 Place Eugène Bataillon 34095 Montpellier Cedex 5 France

Personal website: www.fabiencondamine.org http://www.fabiencondamine.org

Le mar. 1 déc. 2020 à 23:08, bastianheimburger notifications@github.com a écrit :

Hi Fabien,

Have you tried this model with other values of lambda and mu? Perhaps the K will change accordingly.

Indeed, I tried also the default values. Here are the results https://github.com/FabienCondamine/Diversification_analyses/files/5625710/default_ddd.txt. But as you can see, for total_richness=250 the K value and AICc value are as strange as above. Tomorrow I will try other initial values (lambda = 0.01 and mu = 0.001), and maybe I am trying another value for total_richness.

And also, why are you changing the total_richness? This must be set to the total number of species in your clade. Have you an uncertainty about the species number in your clade?

Yes, it is totally unknown how many species there are. That's why, I'm testing three different assumptions in regards to the proportion of missing species within the clade/phylogenetic tree: I have used 114 species (the presumed number of species, but definitely way too low), 150 species (again, too low), and 250 species (more likely but maybe too high). And as such it seems that the DDX model seems to fit best compared over all assumptions. 150 and 250 species are based on results of species delimitations using ~650 specimens.

Cheers Bastian

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/FabienCondamine/Diversification_analyses/issues/1#issuecomment-736851236, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABT5APHDRB3HW32QDC7O2S3SSVSPRANCNFSM4TQKWQDA .

[bheimbu]

Hi,

I have changed the initial values of lambda (0.01) and mu (0.001) -- it got even worse, see here.

Cheers Bastian

[FabienCondamine]

Hey,

You need to try other values, again and again. It can be that weird. Can you run more tests with different values, not just another couple of values but more?

Sorry, but I'm not the one who made the DDD package so perhaps this could be beyond my expertise.

Cheers,

Fabien

Fabien L. Condamine, Ph.D.

CNRS http://www.cnrs.fr/index.php/en, UMR 5554 Institut des Sciences de l'Evolution de Montpellier http://www.isem.univ-montp2.fr/en/ Team Phylogeny and Molecular Evolution https://scholar.google.fr/citations?user=i_bc4KQAAAAJ&hl=en *Université de Montpellier, *Bât. 22 RDC, CC 064 Place Eugène Bataillon 34095 Montpellier Cedex 5 France

Personal website: www.fabiencondamine.org http://www.fabiencondamine.org

Le mer. 2 déc. 2020 à 09:56, bastianheimburger notifications@github.com a écrit :

Hi,

I have changed the initial values of lambda (0.01) and mu (0.001) -- it got even worse, see here https://github.com/FabienCondamine/Diversification_analyses/files/5628018/lambdamu_ddd.txt .

Cheers Bastian

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/FabienCondamine/Diversification_analyses/issues/1#issuecomment-737089127, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABT5APD7KPQNPHYOLFQ7I5LSSX6MXANCNFSM4TQKWQDA .

[bheimbu]

Hi Fabien,

of course I can, but I am not so sure about which values to try. Anyway, thanks for all your help again. I'll get back to you with new results asap.

Cheers Bastian

[FabienCondamine]

Hi,

Try lambda = 0.2 / 0.1 / 0.05 / 0.02; and mu = 0.05 / 0.02 / 0.01

Don't know which combination will work better...

Fabien

Fabien L. Condamine, Ph.D.

CNRS http://www.cnrs.fr/index.php/en, UMR 5554 Institut des Sciences de l'Evolution de Montpellier http://www.isem.univ-montp2.fr/en/ Team Phylogeny and Molecular Evolution https://scholar.google.fr/citations?user=i_bc4KQAAAAJ&hl=en *Université de Montpellier, *Bât. 22 RDC, CC 064 Place Eugène Bataillon 34095 Montpellier Cedex 5 France

Personal website: www.fabiencondamine.org http://www.fabiencondamine.org

Le mer. 2 déc. 2020 à 10:04, bastianheimburger notifications@github.com a écrit :

Hi Fabien,

of course I can, but I am not so sure about which values to try. Anyway, thanks for all your help again. I'll get back to you with new results asap.

Cheers Bastian

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/FabienCondamine/Diversification_analyses/issues/1#issuecomment-737093113, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABT5APBQ4VLZVI7ZESKKAI3SSX7ITANCNFSM4TQKWQDA .

[bheimbu]

Hi Fabien,

it took a while to try various combinations -- and I didnt find the best-fitting one. I have used following combinations (lambda/mu): 0.3/0.2, 0.3/0.1, 0.2/0.05, 0.1/0.05. And I tried different values for DDL+E and DDX+E here. In the latter, everything looks OK, I would see -- apart from K, which is really skyrocketing again for total_richness=150 and 250?!

Cheers Bastian

[FabienCondamine]

Hey,

Sorry it took so long for you. But now you have a better idea of the models' behavior :-)

Honestly, the last text file you send with a summary of the model estimates is really good according to me. It makes sense that the K value is super high (see Condamine et al. 2019 - Ecol. Lett.), which would mean that diversity dependence is not so strong after all.

I would go with these starting values. I hope this helps.

All the best,

Fabien

Fabien L. Condamine, Ph.D.

CNRS http://www.cnrs.fr/index.php/en, UMR 5554 Institut des Sciences de l'Evolution de Montpellier http://www.isem.univ-montp2.fr/en/ Team Phylogeny and Molecular Evolution https://scholar.google.fr/citations?user=i_bc4KQAAAAJ&hl=en *Université de Montpellier, *Bât. 22 RDC, CC 064 Place Eugène Bataillon 34095 Montpellier Cedex 5 France

Personal website: www.fabiencondamine.org http://www.fabiencondamine.org

Le lun. 7 déc. 2020 à 09:48, bastianheimburger notifications@github.com a écrit :

Hi Fabien,

it took a while to try various combinations -- and I didnt find the best-fitting one. I have used following combinations (lambda/mu): 0.3/0.2 https://github.com/FabienCondamine/Diversification_analyses/files/5651450/0.3_0.2.txt, 0.3/0.1 https://github.com/FabienCondamine/Diversification_analyses/files/5651451/0.3_0.1.txt, 0.2/0.05 https://github.com/FabienCondamine/Diversification_analyses/files/5651452/0.2_0.05.txt, 0.1/0.05 https://github.com/FabienCondamine/Diversification_analyses/files/5651454/0.1_0.05.txt. And I tried different values for DDL+E and DDX+E here https://github.com/FabienCondamine/Diversification_analyses/files/5651470/ddl_ddx.txt. In the latter, everything looks OK, I would see -- apart from K, which is really skyrocketing again for total_richness=150 and 250?!

Cheers Bastian

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/FabienCondamine/Diversification_analyses/issues/1#issuecomment-739771080, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABT5APABRRFTT4SSMZJYVGLSTSJFLANCNFSM4TQKWQDA .

[bheimbu]

Hi Fabien,

ok, if I understand you correct and with respect to your abovementioned paper, the analysis says that a linear diversity-dependent model fits best but the effect is not really strong since it could also be related to changes in temperature, right?

I am really a newbie to diversification anaylsis, thats why I would like to ask another question: How can I justify the changed initial values? It seems a bit arbitrary what I have done but actually those last used values for lambda and mu are based more or less on BAMM results.

Cheers and again thanks for all your help,

Bastian

[FabienCondamine]

Hey,

I would conclude that indeed. If you find that a DDD model is the best fit after comparing it to constant-rate models and time-dependent models for instance, then the estimate of K tells you how strong is the diversity dependence in the focal clade. So you can perhaps measure the difference between K and the total richness of your clade (see Condamine 2018 - Biology Letters https://royalsocietypublishing.org/doi/full/10.1098/rsbl.2017.0622).

No problem. It's good to learn a lot and when you start in a field, I guess we are all newbies ;-) The use of different starting values is just to ensure good convergence in the parameter estimation. Sometimes, the logL is very different from one initial value to the other in ML analyses perhaps due to local optima in the likelihood landscape. So it's important to test for different values and find the same (similar) results. These values are standard in the field, so don't worry so much about that. BAMM is a Bayesian approach so there are many posterior estimates of the values, not like in ML where you have a point estimate...

I hope this helps ;-) Best regards,

Fabien

Fabien L. Condamine, Ph.D.

CNRS http://www.cnrs.fr/index.php/en, UMR 5554 Institut des Sciences de l'Evolution de Montpellier http://www.isem.univ-montp2.fr/en/ Team Phylogeny and Molecular Evolution https://scholar.google.fr/citations?user=i_bc4KQAAAAJ&hl=en *Université de Montpellier, *Bât. 22 RDC, CC 064 Place Eugène Bataillon 34095 Montpellier Cedex 5 France

Personal website: www.fabiencondamine.org http://www.fabiencondamine.org

Le lun. 7 déc. 2020 à 14:46, bastianheimburger notifications@github.com a écrit :

Hi Fabien,

ok, if I understand you correct and with respect to your abovementioned paper, the analysis says that a linear diversity-dependent model fits best but the effect is not really strong since it could also be related to changes in temperature, right?

I am really a newbie to diversification anaylsis, thats why I would like to ask another question: How can I justify the changed initial values? It seems a bit arbitrary what I have done but actually those last used values for lambda and mu are based more or less on BAMM results.

Cheers and again thanks for all your help,

Bastian

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/FabienCondamine/Diversification_analyses/issues/1#issuecomment-739928027, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABT5APAKCNJFXXECK4TVR4DSTTMEFANCNFSM4TQKWQDA .

[bheimbu]

Hi Fabien,

busy times, beg my pardon for not reaching back to you earlier. I will try your suggested approach about K and the strength of the diversity dependence asap.

Regards,

Bastian

Sent from my iPhone

On 7. Dec 2020, at 18:46, Fabien Condamine notifications@github.com wrote:



Hey,

I would conclude that indeed. If you find that a DDD model is the best fit after comparing it to constant-rate models and time-dependent models for instance, then the estimate of K tells you how strong is the diversity dependence in the focal clade. So you can perhaps measure the difference between K and the total richness of your clade (see Condamine 2018 - Biology Letters https://royalsocietypublishing.org/doi/full/10.1098/rsbl.2017.0622).

No problem. It's good to learn a lot and when you start in a field, I guess we are all newbies ;-) The use of different starting values is just to ensure good convergence in the parameter estimation. Sometimes, the logL is very different from one initial value to the other in ML analyses perhaps due to local optima in the likelihood landscape. So it's important to test for different values and find the same (similar) results. These values are standard in the field, so don't worry so much about that. BAMM is a Bayesian approach so there are many posterior estimates of the values, not like in ML where you have a point estimate...

I hope this helps ;-) Best regards,

Fabien

Fabien L. Condamine, Ph.D.

CNRS http://www.cnrs.fr/index.php/en, UMR 5554 Institut des Sciences de l'Evolution de Montpellier http://www.isem.univ-montp2.fr/en/ Team Phylogeny and Molecular Evolution https://scholar.google.fr/citations?user=i_bc4KQAAAAJ&hl=en *Université de Montpellier, *Bât. 22 RDC, CC 064 Place Eugène Bataillon 34095 Montpellier Cedex 5 France

Personal website: www.fabiencondamine.org http://www.fabiencondamine.org

Le lun. 7 déc. 2020 à 14:46, bastianheimburger notifications@github.com a écrit :

Hi Fabien,

ok, if I understand you correct and with respect to your abovementioned paper, the analysis says that a linear diversity-dependent model fits best but the effect is not really strong since it could also be related to changes in temperature, right?

I am really a newbie to diversification anaylsis, thats why I would like to ask another question: How can I justify the changed initial values? It seems a bit arbitrary what I have done but actually those last used values for lambda and mu are based more or less on BAMM results.

Cheers and again thanks for all your help,

Bastian

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/FabienCondamine/Diversification_analyses/issues/1#issuecomment-739928027, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABT5APAKCNJFXXECK4TVR4DSTTMEFANCNFSM4TQKWQDA .

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHubhttps://github.com/FabienCondamine/Diversification_analyses/issues/1#issuecomment-740074047, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AF67TIYG62BRBZYPACN67HDSTUIIXANCNFSM4TQKWQDA.