Temporal signal with Phytest

In this example we use [phytest]() to enure that there is temporal signal in a dataset of sequences sampled through time. Temporal signal analysis is an important step for detecting problematic sequences and potential issues before heading on to a Bayesian phylogenetic analysis (e.g. with BEAST).

This example uses data from the TempEst tutorial. TempEst is a useful program for explore a temporal signal analysis though a GUI, however, it is not possible to automate TempEst. Under the hood Phytest uses TimeTree to perform root-to-tip regression allowing us to automate temporal signal testing. Automating this process is especially useful when an analysis is run many times e.g. through development or during daily builds.

Setup

Create a python environment

python3 -m venv venv

Activate the environment

source venv/bin/activate

Install phytest with pip

python -m pip install phytest

Tests

In the ice_viruses_tests.py file we define the tests that we want to run on our dataset. Phytest is a flexible framework that provides many built in tests and is easily extendable. For simplicity we will focus on a small test set but see the phytest docs for more information.

The focus of this test will be determining if the dataset meets our temporal signal requirements. in the file we file the test below. We want the r squared value of our root-to-tip regression to be greater that 0.5 (this suggest a strong correlation between branch length and sampling date), we want the inferred rate to be between 0.002 and 0.005 (the evolutionary rate of influenza is on the order of 2.8–4.4 × 10^–3), and finally we expect the root date i.e. TMRCA to be in the 19th century.

def test_root_to_tip_r_squared(tree: Tree):
    tree.assert_root_to_tip(min_r_squared=0.5)

def test_root_to_tip_rate(tree: Tree):
    tree.assert_root_to_tip(min_rate=0.002, max_rate=0.005)

def test_root_to_tip_date(tree: Tree):
    tree.assert_root_to_tip(min_root_date=1800, max_root_date=1900)

If any of these test fail then then phytest will fail. While these test aren't very strict (we give ourselves a lot of wiggle room) they are still useful for picking up errors before the analysis is run.

The root-to-tip regression can be plotted with the tree.plot_root_to_tip methods e.g. tree.plot_root_to_tip("root-to-tip.png").

Data

Here we test two data sets from the TempEst tutorial. The first is an uncleaned dataset that contains erroneous data and the seconds is a cleaned dataset that has the sequences removed.

Uncleaned

phytest ice_viruses_tests.py -s data/ice_viruses.fasta -t data/ice_viruses.fasta.treefile

From the output we can see that our tests failed.

Results (0.65s):
       3 failed
         - ice_viruses_tests.py:4 test_root_to_tip_r_squared[tree0]
         - ice_viruses_tests.py:8 test_root_to_tip_rate[tree0]
         - ice_viruses_tests.py:12 test_root_to_tip_date[tree0]

Using the --report report.html flag we can create a detailed report of results.

Cleaned

phytest ice_viruses_tests.py -s data/ice_viruses_cleaned.fasta -t data/ice_viruses_cleaned.fasta.treefile

Using the cleaned dataset all our tests pass, giving us confidence to progress with our analysis.

Results (0.60s):
      3 passed