Selecting models for phylogenetic analysis

We use the piq_model_finder app to rank models. This is the python binding to the IQ-TREE ModelFinder tool.

In [1]:

from cogent3 import app_help, get_app, load_aligned_seqs

from piqtree import download_dataset

aln_path = download_dataset("example.phy.gz", dest_dir="data")
aln = load_aligned_seqs(aln_path, moltype="dna", format_name="phylip")

from cogent3 import app_help, get_app, load_aligned_seqs from piqtree import download_dataset aln_path = download_dataset("example.phy.gz", dest_dir="data") aln = load_aligned_seqs(aln_path, moltype="dna", format_name="phylip")

Get help and then apply piq_model_finder.

In [2]:

app_help("piq_model_finder")

app_help("piq_model_finder")

Overview
--------
Find the models of best fit for an alignment using ModelFinder.

Options for making the app
--------------------------
piq_model_finder_app = get_app('piq_model_finder', *args, **kwargs)

Parameters
----------
aln : Alignment
    The alignment to find the model of best fit for.
model_set : Iterable[str] | None, optional
    Search space for models.
    Equivalent to IQ-TREE's mset parameter, by default None
freq_set : Iterable[str] | None, optional
    Search space for frequency types.
    Equivalent to IQ-TREE's mfreq parameter, by default None
rate_set : Iterable[str] | None, optional
    Search space for rate heterogeneity types.
    Equivalent to IQ-TREE's mrate parameter, by default None
rand_seed : int | None, optional
    The random seed - None means no seed is used, by default None.
num_threads: int | None, optional
    Number of threads for IQ-TREE to use, by default None (single-threaded).
    If 0 is specified, IQ-TREE attempts to find the optimal number of threads.
other_options: str, optional
    Additional command line options for IQ-TREE.

Returns
-------
ModelFinderResult
    Collection of data returned from IQ-TREE's ModelFinder.

Input type
----------
Alignment

Output type
-----------
ModelFinderResult

In [3]:

mfinder = get_app("piq_model_finder")
ranked = mfinder(aln)
ranked

mfinder = get_app("piq_model_finder") ranked = mfinder(aln) ranked

Out[3]:

ModelFinderResult(source=None, best_aic=Model(submod_type=GTR, freq_type=F, rate_type=I+G4), best_aicc=Model(submod_type=GTR, freq_type=F, rate_type=I+G4), best_bic=Model(submod_type=TIM2, freq_type=F, rate_type=I+G4))

Accessing the best model¶

The different measures used to select the best model, AIC, AICc, and BIC, are available as attributes of the result object. We'll select AICc as the measure for choosing the best model.

In [4]:

selected = ranked.best_aicc

selected = ranked.best_aicc

You can inspect the statistics for one of these using the model_stats attribute.

In [5]:

ranked.model_stats[selected]

ranked.model_stats[selected]

Out[5]:

ModelResultValue(lnL=-21148.95667, nfp=41, tree_length=4.200251662)

Using the best model¶

You can apply the selected model to a phylogenetic analysis.

Note The process is the same for both the piq_build_tree and the piq_fit_tree apps.

In [6]:

fit = get_app("piq_build_tree", selected)
fitted = fit(aln)
fitted

fit = get_app("piq_build_tree", selected) fitted = fit(aln) fitted

Out[6]:

Tree("(LngfishAu,(LngfishSA,LngfishAf),(Frog,((((Turtle,(Crocodile,Bird)),Sphenodon),Lizard),(((Human,(Seal,(Cow,Whale))),(Mouse,Rat)),(Platypus,Opossum)))));")