Questions about BUSTED and an error code 102 about hyphy

[spond]

Dear @irislin886,

I can't replicate the issue locally, i.e. when I run the analysis using hyphy (develop branch), everything finishes. The error occurs (for the version on Datamonkey) because the background branches have dN/dS = 0. This will be fixed in the next version update.

### Performing the full (dN/dS > 1 allowed) branch-site model fit
* Log(L) =  -736.37, AIC-c =  1545.65 (34 estimated parameters)
* For *test* branches, the following rate distribution for branch-site combinations was inferred

|          Selection mode           |     dN/dS     |Proportion, %|               Notes               |
|-----------------------------------|---------------|-------------|-----------------------------------|
|        Negative selection         |     0.000     |   19.178    |                                   |
|        Negative selection         |     0.000     |   80.822    |       Collapsed rate class        |
|      Diversifying selection       |     3.578     |    0.000    |       Not supported by data       |

* For *background* branches, the following rate distribution for branch-site combinations was inferred

|          Selection mode           |     dN/dS     |Proportion, %|               Notes               |
|-----------------------------------|---------------|-------------|-----------------------------------|
|        Negative selection         |     0.029     |   49.170    |                                   |
|        Negative selection         |     0.029     |   41.964    |       Collapsed rate class        |
|      Diversifying selection       |     1.294     |    8.866    |                                   |

* The following rate distribution for site-to-site **synonymous** rate variation was inferred

|               Rate                | Proportion, % |               Notes               |
|-----------------------------------|---------------|-----------------------------------|
|               0.663               |    84.061     |                                   |
|               0.667               |     8.531     |       Collapsed rate class        |
|               5.204               |     7.408     |                                   |


### No evidence for episodic diversifying positive selection under the unconstrained model, skipping constrained model fitting
----
## Branch-site unrestricted statistical test of episodic diversification [BUSTED]
Likelihood ratio test for episodic diversifying positive selection, **p =   0.5000**.

As a general rule of thumb, very small alignments (your has 4 sequences and 130 codons, 0.2 total tree length) often show unstable behavior with more complex models like BUSTED, simply because there's not enough information to fit the model. There's also an extreme imbalance in branch lengths. In the following tree, the labels show the estimated NUMBER of substitutions (syn + non-syn) along each branch. You can see that for most branches its ~1.

Personally, for something this small, I'd use a plain old FitMG94 analysis (https://github.com/veg/hyphy-analyses/tree/master/FitMG94)

hyphy ~/Development/hyphy-analyses/FitMG94/FitMG94.bf --alignment ~/Desktop/test.txt --lrt Yes --type local

|            Branch            |     Length     |     dN/dS      |Approximate dN/dS CI|LRT p-value dN != dS|
|:----------------------------:|:--------------:|:--------------:|:------------------:|:------------------:|
|ANA_STRENUUS_G39463_T2_GALB...|     0.002      |     0.000      |   0.000 - 0.535    |       0.0883       |
|ANAX_WALSINGHAMI_G29729_T2_...|     0.002      |     0.000      |   0.000 - 0.533    |       0.0881       |
|            Node3             |     0.008      |     0.000      |   0.000 - 0.174    |       0.1264       |
|ANAX_JUNIUS_G42871_T2_GALBA...|     0.000      |     1.000      |0.000 - 10000.000...|       0.9974       |
|TANYPTERYX_HAGENI_G42552_T2...|     0.274      |     0.112      |   0.076 - 0.161    |       0.0000       |

which shows that only the longest branch TANYPTERYX_HAGENI_G42552_T2... has any evidence for dN/dS ≠ 1 (and even then it's because dN/dS < 1).

Best,
Sergei

[spond]

Dear @irislin886,

This is a configuration issue on our end. Do you have a URL for the failed job?

Best,
Sergei

@stevenweaver

[irislin886]

Yes, I do! here is the url: https://www.datamonkey.org/busted/6627f2dacc6c1a3e32b6732c

[stevenweaver]

That message is actually a bit misleading (but should be resolved). The issue is the following, which can be found by clicking the Download Log button:

### Performing the full (dN/dS > 1 allowed) branch-site model fit
Error:
busted.test.omega1 evaluated to a NaN; this can cause all kinds of odd behavior downstream, therefore it is safer to quit now

Function call stack
1 :  [namespace = SloudcTB] Optimize(mles, likelihoodFunction, run_options[utility.getGlobalValue("terms.run_options.optimization_settings")]);

        Keyword arguments:
                {
                 "save-fit":"/dev/null"
                }
-------
2 :  busted.grid_search.results=estimators.FitLF(busted.filter_names,busted.trees,busted.model_map,busted.final_partitioned_mg_results,busted.model_object_map,{terms.run_options.retain_lf_object:TRUE,terms.run_options.proportional_branch_length_scaler:busted.global_scaler_list,terms.run_options.optimization_settings:{"OPTIMIZATION_METHOD":"nedler-mead","MAXIMUM_OPTIMIZATION_ITERATIONS":500,"OPTIMIZATION_PRECISION":busted.nm.precision},terms.search_grid:busted.initial_grid,terms.search_restarts:busted.N.initial_guesses});

        Keyword arguments:
                {
                 "save-fit":"/dev/null"
                }
-------
                                                 

Best,
Steven

[stevenweaver]

The conflict message should no longer be appearing (again, this has no bearing on the actual outcome of the job).

Best,
Steven

[irislin886]

Thank you very much, Steven and Sergei.

However, after reading the error message from the log, I don't quite understand what is happening. Would you mind explaining the actual issue to me and how I can fix it?

Best,
Iris

[irislin886]

Dear Sergi,

Thank you very much for your help and the explanation. I ran it locally using FitMG94 and it works great. Thank you very much for the suggestions.

Best,
Iris