Merge pull request #413 from monarch-initiative/develop

ielis · web-flow · commit 1b1bb5e648c5 · 2025-01-31T17:29:52.000+01:00
Release `v0.9.4`
diff --git a/docs/conf.py b/docs/conf.py
@@ -64,7 +64,7 @@
 # The short X.Y version.
 version = u'0.9'
 # The full version, including alpha/beta/rc tags.
-release = u'0.9.3'
+release = u'0.9.4'
 
 # The language for content autogenerated by Sphinx. Refer to documentation
 # for a list of supported languages.
diff --git a/docs/tutorial.rst b/docs/tutorial.rst
@@ -314,7 +314,7 @@ For general use, we recommend using a combination
 of a *phenotype MT filter* (:class:`~gpsea.analysis.mtc_filter.PhenotypeMtcFilter`) with a *multiple testing correction*.
 Phenotype MT filter chooses the HPO terms to test according to several heuristics, which
 reduce the multiple testing burden and focus the analysis
-on the most interesting terms (see :ref:`HPO MT filter <hpo-mt-filter>` for more info).
+on the most interesting terms (see :ref:`Independent filtering for HPO <hpo-if-filter>` for more info).
 Then the multiple testing correction, such as Bonferroni or Benjamini-Hochberg,
 is used to control the family-wise error rate or the false discovery rate.
 See :ref:`mtc` for more information.
@@ -323,7 +323,7 @@ See :ref:`mtc` for more information.
 >>> analysis = configure_hpo_term_analysis(hpo)
 
 :func:`~gpsea.analysis.pcats.configure_hpo_term_analysis` configures the analysis
-that uses HPO MTC filter (:class:`~gpsea.analysis.mtc_filter.HpoMtcFilter`) for selecting HPO terms of interest,
+that uses Independent filtering for HPO (:class:`~gpsea.analysis.mtc_filter.IfHpoFilter`) for selecting HPO terms of interest,
 Fisher Exact test for computing nominal p values, and Benjamini-Hochberg for multiple testing correction.
 
 
diff --git a/docs/user-guide/analyses/mtc.rst b/docs/user-guide/analyses/mtc.rst
@@ -171,31 +171,37 @@ we pass an iterable (e.g. a tuple) with these two terms as an argument:
 2
 
 
-.. _hpo-mt-filter:
+.. _hpo-if-filter:
 
-HPO MT filter
--------------
+Independent filtering for HPO
+-----------------------------
+
+Independent filtering for HPO involves making several domain judgments
+and taking advantage of the HPO structure
+in order to reduce the number of HPO terms for testing.
+The filter's logic is made up of 8 individual heuristics 
+to skip testing the terms that are unlikely to yield significant or interesting results (see below).
 
-The HPO MT filter involves making several domain judgments and takes advantage of the HPO structure.
-The strategy needs access to HPO:
+Some of the heuristics need to access HPO hierarchy,
+so let's load HPO
 
 >>> import hpotk
 >>> store = hpotk.configure_ontology_store()
 >>> hpo = store.load_minimal_hpo(release='v2024-07-01')
 
-and it is implemented in the :class:`~gpsea.analysis.mtc_filter.HpoMtcFilter` class:
+and let's create the  :class:`~gpsea.analysis.mtc_filter.IfHpoFilter` class
+using the static constructor
+:func:`~gpsea.analysis.mtc_filter.IfHpoFilter.default_filter`:
 
->>> from gpsea.analysis.mtc_filter import HpoMtcFilter
->>> hpo_mtc = HpoMtcFilter.default_filter(hpo=hpo)
+>>> from gpsea.analysis.mtc_filter import IfHpoFilter
+>>> hpo_mtc = IfHpoFilter.default_filter(hpo=hpo)
 
 
-We use static constructor :func:`~gpsea.analysis.mtc_filter.HpoMtcFilter.default_filter`
-for creating :class:`~gpsea.analysis.mtc_filter.HpoMtcFilter`.
-The constructor takes a ``term_frequency_threshold`` option (40% by default) 
-and the method's logic is made up of 8 individual heuristics 
-designed to skip testing the HPO terms that are unlikely to yield significant or interesting results.
+The constructor takes HPO and two thresholds (optional).
+See the API documentation and the explanations below for more details.
 
-.. contents:: HPO MT filters
+
+.. contents:: Independent filtering for HPO
   :depth: 1
   :local:
 
@@ -281,6 +287,8 @@ that if there is a signal from the nervous system,
 it will lead to at least one of the descendents of
 *Abnormality of the nervous system* being significant.
 
+See :ref:`general-hpo-terms` section for details.
+
 
 `HMF09` - Skipping terms that are rare on the cohort level 
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
@@ -294,8 +302,6 @@ and we have explicit observed observations for 20 and excluded for 10 individual
 then the annotation frequency is `0.3`. 
 
 The threshold is set as ``annotation_frequency_threshold`` option
-of the :func:`~gpsea.analysis.mtc_filter.HpoMtcFilter.default_filter` constructor,
+of the :func:`~gpsea.analysis.mtc_filter.IfHpoFilter.default_filter` constructor,
 with the default value of `0.4` (40%).
 
-
-See :ref:`general-hpo-terms` section for details.
diff --git a/docs/user-guide/analyses/phenotype-classes.rst b/docs/user-guide/analyses/phenotype-classes.rst
@@ -207,7 +207,7 @@ a phenotype multiple testing (MT) filter and multiple testing correction (MTC).
 
 Phenotype MT filter selects a (sub)set of HPO terms for testing,
 for instance only the user-selected terms (see :class:`~gpsea.analysis.mtc_filter.SpecifiedTermsMtcFilter`)
-or the terms selected by :class:`~gpsea.analysis.mtc_filter.HpoMtcFilter`.
+or the terms selected by :class:`~gpsea.analysis.mtc_filter.IfHpoFilter`.
 
 MTC then adjusts the nominal p values for the increased risk
 of false positive G/P associations.
@@ -221,8 +221,8 @@ We must choose a phenotype MT filter as well as a MTC procedure to perform genot
 Default analysis
 ^^^^^^^^^^^^^^^^
 
-We recommend using HPO MT filter (:class:`~gpsea.analysis.mtc_filter.HpoMtcFilter`) as a phenotype MT filter
-and Benjamini-Hochberg for MTC.
+We recommend using Independent filtering for HPO (:class:`~gpsea.analysis.mtc_filter.IfHpoFilter`)
+and Benjamini-Hochberg MT correction.
 The default analysis can be configured with :func:`~gpsea.analysis.pcats.configure_hpo_term_analysis` convenience method.
 
 >>> from gpsea.analysis.pcats import configure_hpo_term_analysis
@@ -240,10 +240,10 @@ Custom analysis
 If the default selection of phenotype MT filter and multiple testing correction is not an option,
 we can configure the analysis manually.
 
-First, we choose a phenotype MT filter (e.g. :class:`~gpsea.analysis.mtc_filter.HpoMtcFilter`):
+First, we choose a phenotype MT filter (e.g. :class:`~gpsea.analysis.mtc_filter.IfHpoFilter`):
 
->>> from gpsea.analysis.mtc_filter import HpoMtcFilter
->>> mtc_filter = HpoMtcFilter.default_filter(hpo, term_frequency_threshold=.2)
+>>> from gpsea.analysis.mtc_filter import IfHpoFilter
+>>> mtc_filter = IfHpoFilter.default_filter(hpo, term_frequency_threshold=.2)
 
 .. note::
 
diff --git a/src/gpsea/__init__.py b/src/gpsea/__init__.py
@@ -2,7 +2,7 @@
 GPSEA is a library for finding genotype-phenotype associations.
 """
 
-__version__ = "0.9.3"
+__version__ = "0.9.4"
 
 _overwrite = False
 """
diff --git a/src/gpsea/analysis/mtc_filter/__init__.py b/src/gpsea/analysis/mtc_filter/__init__.py
@@ -6,9 +6,15 @@
 """
 
 from ._impl import PhenotypeMtcFilter, PhenotypeMtcResult, PhenotypeMtcIssue
-from ._impl import UseAllTermsMtcFilter, SpecifiedTermsMtcFilter, HpoMtcFilter
+from ._impl import UseAllTermsMtcFilter, SpecifiedTermsMtcFilter, IfHpoFilter
+from ._impl import HpoMtcFilter
 
 __all__ = [
-    'PhenotypeMtcFilter', 'PhenotypeMtcResult', 'PhenotypeMtcIssue',
-    'UseAllTermsMtcFilter', 'SpecifiedTermsMtcFilter', 'HpoMtcFilter',
+    "PhenotypeMtcFilter",
+    "PhenotypeMtcResult",
+    "PhenotypeMtcIssue",
+    "UseAllTermsMtcFilter",
+    "SpecifiedTermsMtcFilter",
+    "IfHpoFilter",
+    "HpoMtcFilter",
 ]
diff --git a/src/gpsea/analysis/mtc_filter/_impl.py b/src/gpsea/analysis/mtc_filter/_impl.py
@@ -3,6 +3,7 @@
 import typing
 
 from collections import deque
+import warnings
 
 import hpotk
 import pandas as pd
@@ -252,14 +253,14 @@ def verify_term_id(val: typing.Union[str, hpotk.TermId]) -> hpotk.TermId:
             raise ValueError(f"{val} is neither `str` nor `hpotk.TermId`")
 
 
-class HpoMtcFilter(PhenotypeMtcFilter[hpotk.TermId]):
+class IfHpoFilter(PhenotypeMtcFilter[hpotk.TermId]):
     """
-    `HpoMtcFilter` decides which phenotypes should be tested and which phenotypes are not worth testing.
+    `IfHpoFilter` decides which phenotypes should be tested and which phenotypes are not worth testing.
 
     The class leverages a number of heuristics and domain decisions.
-    See :ref:`hpo-mt-filter` section for more info.
+    See :ref:`hpo-if-filter` section for more info.
 
-    We recommend creating an instance using the :func:`default_filter` static factory method.
+    We recommend creating an instance using the :func:`~gpsea.analysis.mtc_filter.IfHpoFilter.default_filter` static factory method.
     """
 
     NO_GENOTYPE_HAS_MORE_THAN_ONE_HPO = PhenotypeMtcResult.fail(
@@ -293,7 +294,7 @@ def default_filter(
               (e.g., 22% in missense and 3% in nonsense genotypes would be OK,
               but not 13% missense and 10% nonsense genotypes if the threshold is 0.2).
               The default threshold is `0.4` (40%).
-            annotation_frequency_threshold: a `float` in range :math:`(0, 1) with the minimum frequency of
+            annotation_frequency_threshold: a `float` in range :math:`(0, 1]` with the minimum frequency of
                 annotation in the cohort. For instance, if the cohort consists of 100 individuals, and
                 we have explicit observed observations for 20 and excluded for 10 individuals, then the
                 annotation frequency is `0.3`. The purpose of this threshold is to omit terms for which
@@ -340,7 +341,7 @@ def default_filter(
         general_hpo_term_set.update(second_level_terms)
         general_hpo_term_set.update(third_level_terms)
 
-        return HpoMtcFilter(
+        return IfHpoFilter(
             hpo=hpo,
             term_frequency_threshold=term_frequency_threshold,
             annotation_frequency_threshold=annotation_frequency_threshold,
@@ -355,7 +356,15 @@ def __init__(
         general_hpo_terms: typing.Iterable[hpotk.TermId],
     ):
         self._hpo = hpo
+        assert (
+            isinstance(term_frequency_threshold, (int, float))
+            and 0.0 < term_frequency_threshold <= 1.0
+        ), "The term_frequency_threshold must be in the range (0, 1]"
         self._hpo_term_frequency_filter = term_frequency_threshold
+        assert (
+            isinstance(annotation_frequency_threshold, (int, float))
+            and 0.0 < annotation_frequency_threshold <= 1.0
+        ), "The annotation_frequency_threshold must be in the range (0, 1]"
         self._hpo_annotation_frequency_threshold = annotation_frequency_threshold
 
         self._general_hpo_terms = set(general_hpo_terms)
@@ -423,17 +432,17 @@ def filter(
                 continue
 
             if term_id in self._general_hpo_terms:
-                results[idx] = HpoMtcFilter.SKIPPING_GENERAL_TERM
+                results[idx] = IfHpoFilter.SKIPPING_GENERAL_TERM
                 continue
 
             if not self._hpo.graph.is_ancestor_of(PHENOTYPIC_ABNORMALITY, term_id):
-                results[idx] = HpoMtcFilter.SKIPPING_NON_PHENOTYPE_TERM
+                results[idx] = IfHpoFilter.SKIPPING_NON_PHENOTYPE_TERM
                 continue
 
             ph_clf = pheno_clfs[idx]
             contingency_matrix = counts[idx]
 
-            max_freq = HpoMtcFilter.get_maximum_group_observed_HPO_frequency(
+            max_freq = IfHpoFilter.get_maximum_group_observed_HPO_frequency(
                 contingency_matrix,
                 ph_clf=ph_clf,
             )
@@ -459,19 +468,19 @@ def filter(
                 results[idx] = self._not_powered_for_2_by_3
                 continue
 
-            if not HpoMtcFilter.some_cell_has_greater_than_one_count(
+            if not IfHpoFilter.some_cell_has_greater_than_one_count(
                 counts=contingency_matrix,
                 ph_clf=ph_clf,
             ):
-                results[idx] = HpoMtcFilter.NO_GENOTYPE_HAS_MORE_THAN_ONE_HPO
+                results[idx] = IfHpoFilter.NO_GENOTYPE_HAS_MORE_THAN_ONE_HPO
                 continue
 
-            elif HpoMtcFilter.one_genotype_has_zero_hpo_observations(
+            elif IfHpoFilter.one_genotype_has_zero_hpo_observations(
                 counts=contingency_matrix,
                 gt_clf=gt_clf,
             ):
                 results[idx] = (
-                    HpoMtcFilter.SKIPPING_SINCE_ONE_GENOTYPE_HAD_ZERO_OBSERVATIONS
+                    IfHpoFilter.SKIPPING_SINCE_ONE_GENOTYPE_HAD_ZERO_OBSERVATIONS
                 )
                 continue
 
@@ -495,7 +504,7 @@ def filter(
                     axis=None
                 ) < 1:
                     # Do not test if the count is exactly the same to the counts in the only child term.
-                    results[idx] = HpoMtcFilter.SAME_COUNT_AS_THE_ONLY_CHILD
+                    results[idx] = IfHpoFilter.SAME_COUNT_AS_THE_ONLY_CHILD
                     continue
 
             # ##
@@ -520,18 +529,18 @@ def possible_results(self) -> typing.Collection[PhenotypeMtcResult]:
         return (
             PhenotypeMtcFilter.OK,
             self._below_frequency_threshold,  # HMF01
-            HpoMtcFilter.NO_GENOTYPE_HAS_MORE_THAN_ONE_HPO,  # HMF02
-            HpoMtcFilter.SAME_COUNT_AS_THE_ONLY_CHILD,  # HMF03
-            HpoMtcFilter.SKIPPING_SINCE_ONE_GENOTYPE_HAD_ZERO_OBSERVATIONS,  # HMF05
+            IfHpoFilter.NO_GENOTYPE_HAS_MORE_THAN_ONE_HPO,  # HMF02
+            IfHpoFilter.SAME_COUNT_AS_THE_ONLY_CHILD,  # HMF03
+            IfHpoFilter.SKIPPING_SINCE_ONE_GENOTYPE_HAD_ZERO_OBSERVATIONS,  # HMF05
             self._not_powered_for_2_by_2,  # HMF06
             self._not_powered_for_2_by_3,  # HMF06
-            HpoMtcFilter.SKIPPING_NON_PHENOTYPE_TERM,  # HMF07
-            HpoMtcFilter.SKIPPING_GENERAL_TERM,  # HMF08
+            IfHpoFilter.SKIPPING_NON_PHENOTYPE_TERM,  # HMF07
+            IfHpoFilter.SKIPPING_GENERAL_TERM,  # HMF08
             self._below_annotation_frequency_threshold,  # HMF09
         )
 
     def filter_method_name(self) -> str:
-        return "HPO MTC filter"
+        return "Independent filtering HPO filter"
 
     @staticmethod
     def get_number_of_observed_hpo_observations(
@@ -623,3 +632,65 @@ def _get_ordered_terms(
 
         # now, ordered_term_list is ordered from leaves to root
         return ordered_term_list
+
+
+class HpoMtcFilter(IfHpoFilter):
+    """
+    `HpoMtcFilter` is deprecated and will be removed in `1.0.0`.
+
+    Use :class:`gpsea.analysis.mtc_filter.IfHpoFilter` instead.
+    """
+
+    @staticmethod
+    def default_filter(
+        hpo: hpotk.MinimalOntology,
+        term_frequency_threshold: float = 0.4,
+        annotation_frequency_threshold: float = 0.4,
+        phenotypic_abnormality: hpotk.TermId = PHENOTYPIC_ABNORMALITY,
+    ):
+        """
+        Args:
+            hpo: HPO
+            term_frequency_threshold: a `float` in range :math:`(0, 1]` with the minimum frequency
+              for an HPO term to have in at least one of the genotype groups
+              (e.g., 22% in missense and 3% in nonsense genotypes would be OK,
+              but not 13% missense and 10% nonsense genotypes if the threshold is 0.2).
+              The default threshold is `0.4` (40%).
+            annotation_frequency_threshold: a `float` in range :math:`(0, 1]` with the minimum frequency of
+                annotation in the cohort. For instance, if the cohort consists of 100 individuals, and
+                we have explicit observed observations for 20 and excluded for 10 individuals, then the
+                annotation frequency is `0.3`. The purpose of this threshold is to omit terms for which
+                we simply do not have much data overall. By default, we set a threshold to `0.4` (40%).
+            phenotypic_abnormality: a :class:`~hpotk.TermId` corresponding to the root of HPO phenotype hierarchy.
+              Having to specify this option should be very rarely, if ever.
+        """
+        warnings.warn(
+            "HpoMtcFilter has been deprecated and will be removed in 1.0.0. Use `IfHpoFilter` instead.",
+            DeprecationWarning,
+            stacklevel=2,
+        )
+        IfHpoFilter.default_filter(
+            hpo=hpo,
+            term_frequency_threshold=term_frequency_threshold,
+            annotation_frequency_threshold=annotation_frequency_threshold,
+            phenotypic_abnormality=phenotypic_abnormality,
+        )
+
+    def __init__(
+        self,
+        hpo: hpotk.MinimalOntology,
+        term_frequency_threshold: float,
+        annotation_frequency_threshold: float,
+        general_hpo_terms: typing.Iterable[hpotk.TermId],
+    ):
+        super().__init__(
+            hpo,
+            term_frequency_threshold,
+            annotation_frequency_threshold,
+            general_hpo_terms,
+        )
+        warnings.warn(
+            "HpoMtcFilter has been deprecated and will be removed in 1.0.0. Use `IfHpoFilter` instead.",
+            DeprecationWarning,
+            stacklevel=2,
+        )
diff --git a/src/gpsea/analysis/pcats/_config.py b/src/gpsea/analysis/pcats/_config.py
@@ -2,7 +2,7 @@
 
 import hpotk
 
-from ..mtc_filter import HpoMtcFilter
+from ..mtc_filter import IfHpoFilter
 from ._impl import HpoTermAnalysis
 from .stats import CountStatistic, FisherExactTest
 
@@ -16,13 +16,13 @@ def configure_hpo_term_analysis(
     """
     Configure HPO term analysis with default parameters.
 
-    The default analysis will pre-filter HPO terms with :class:`~gpsea.analysis.mtc_filter.HpoMtcFilter`,
+    The default analysis will pre-filter HPO terms with :class:`~gpsea.analysis.mtc_filter.IfHpoFilter`,
     then compute nominal p values using `count_statistic` (default Fisher exact test),
     and apply multiple testing correction (default Benjamini/Hochberg (`fdr_bh`))
     with target `mtc_alpha` (default `0.05`).
     """
     return HpoTermAnalysis(
-        mtc_filter=HpoMtcFilter.default_filter(hpo),
+        mtc_filter=IfHpoFilter.default_filter(hpo),
         count_statistic=count_statistic,
         mtc_correction=mtc_correction,
         mtc_alpha=mtc_alpha,
diff --git a/tests/analysis/pcats/test_hpo_term_analysis.py b/tests/analysis/pcats/test_hpo_term_analysis.py
diff --git a/tests/analysis/test_mtc_filter.py b/tests/analysis/test_mtc_filter.py
