Merge pull request #243 from monarch-initiative/ielis/issue240

Add T test

docs/user-guide/stats.rst

@@ -463,6 +463,12 @@ We will use :ref:`mann-whitney-u-test` as described above.
 >>> score_statistic = MannWhitneyStatistic()
 
 
+.. tip::
+
+   See :mod:`gpsea.analysis.pscore.stats` module for more statistical tests available
+   for using with phenotype scores.
+
+
 **Final analysis**
 
 We will put the final analysis together into :class:`~gpsea.analysis.pscore.PhenotypeScoreAnalysis`.

src/gpsea/analysis/pscore/stats/__init__.py

@@ -1,7 +1,7 @@
 from ._stats import PhenotypeScoreStatistic
-from ._stats import MannWhitneyStatistic
+from ._stats import MannWhitneyStatistic, TTestStatistic
 
 __all__ = [
     'PhenotypeScoreStatistic',
-    'MannWhitneyStatistic',
+    'MannWhitneyStatistic', 'TTestStatistic',
 ]

src/gpsea/analysis/pscore/stats/_stats.py

@@ -1,7 +1,7 @@
 import abc
 import typing
 
-from scipy.stats import mannwhitneyu
+from scipy.stats import mannwhitneyu, ttest_ind
 
 
 class PhenotypeScoreStatistic(metaclass=abc.ABCMeta):
@@ -42,3 +42,25 @@ def compute_pval(
         )
 
         return pval
+
+
+class TTestStatistic(PhenotypeScoreStatistic):
+    """
+    `TTestStatistic` is a wrapper around SciPy's
+    :func:`~scipy.stats.ttest_ind` function to apply
+    T test on 2 phenotype scores.
+    """
+
+    def compute_pval(
+        self,
+        scores: typing.Collection[typing.Sequence[float]],
+    ) -> float:
+        assert len(scores) == 2, 'T test only supports 2 categories at this time'
+
+        x, y = scores
+        res = ttest_ind(
+            a=x, b=y,
+            alternative='two-sided',
+        )
+
+        return res.pvalue

tests/analysis/pscore/test_stats.py

@@ -0,0 +1,54 @@
+import typing
+import pytest
+
+from gpsea.analysis.pscore.stats import MannWhitneyStatistic, TTestStatistic
+
+
+class TestMannWhitneyStatistic:
+
+    @pytest.fixture(scope='class')
+    def statistic(self) -> MannWhitneyStatistic:
+        return MannWhitneyStatistic()
+
+    @pytest.mark.parametrize(
+        'x, y, expected',
+        [
+            ((1., 2., 3., ), (1., 2., 3., ), 1.),
+            ((11., 15, 8., 12.,), (4., 2., 3., 3.5, 4.,), 0.01945103333136247),
+        ]
+    )
+    def test_compute_pval(
+        self,
+        statistic: MannWhitneyStatistic,
+        x: typing.Sequence[float],
+        y: typing.Sequence[float],
+        expected: float,
+    ):
+        actual = statistic.compute_pval((x, y))
+
+        assert actual == pytest.approx(expected)
+
+
+class TestTTestStatistic:
+
+    @pytest.fixture(scope='class')
+    def statistic(self) -> TTestStatistic:
+        return TTestStatistic()
+
+    @pytest.mark.parametrize(
+        'x, y, expected',
+        [
+            ((1., 2., 3., ), (1., 2., 3., ), 1.),
+            ((11., 15, 8., 12.,), (4., 2., 3., 3.5, 4.,), 0.0004749950471148506),
+        ]
+    )
+    def test_compute_pval(
+        self,
+        statistic: TTestStatistic,
+        x: typing.Sequence[float],
+        y: typing.Sequence[float],
+        expected: float,
+    ):
+        actual = statistic.compute_pval((x, y))
+
+        assert actual == pytest.approx(expected)