refactor(L2GFeatureMatrix)!: streamline feature matrix management (op…

…entargets#745)

* refactor(L2GFeatureMatrix): remove schema validation

* refactor(FeatureFactory): reshape feature generation WIP

* chore: pre-commit auto fixes [...]

* chore: set l2gfeature properties with decorator

* chore(l2gfeature): make credible_set and input_dependency instance attributes

* chore(l2gfeature): make credible_set and input_dependency instance attributes

* chore(featurefactory): distanceTssMeanFeature working

* refactor(l2g): improve step dependency management

* feat: implement

* chore: fix mypy issues

* feat: l2gfeaturematrix.from_features_list working

* chore: comment out obsolete refs

* chore(L2GFeatureMatrix): change `mode` attribute to `with_gold_standard`

* refactor(l2g): move feature matrix writing to training module

* feat(L2GFeatureMatrix): accept L2GGoldStandard or StudyLocus as inputs

* feat: implement methods to build a feature matrix based on a studylocus/L2GGoldStandard instance

* feat: coloc logic prototype

* feat(l2g): filter non gwas credible sets at the start of the step

* feat: rewrite colocalisation feature factory

* test: add `test_colocalisation_feature_type`

* test(colocalisation): add test_extract_maximum_coloc_probability_per_region_and_gene

* feat(L2GFeatureInputLoader): support multiple deps by passing loader as kwarg

* test: add integration tests `test_build_feature_matrix`

* chore: drop config yamls

* refactor: move feature classes to datasets module

* docs: update feature docs

* refactor(colocalisation): cleaner joins in `append_right_study_metadata`

* chore: better logging abstract methods

* test: add `L2GFeatureMatrix.test_from_features_list` unit tests

* fix: add goldStandardSet when a gs instance is passed to `from_features_list`

* fix: lowercase colocalisation type and add semantic test

* test: add semantic test for `append_right_study_metadata`

* feat(colocalisation): make `append_right_study_metadata` extensible to left metadata

* fix(colocalisation): append_study_metadata cant take a gold standard

* fix(colocalisation): extract_maximum_coloc_probability_per_region_and_gene cant take a gold standard

* feat: add `StudyLocus` as a dependency of colocalisation features

* fix: add studylocus to input loader in test

* fix: add studylocus to input loader in test

* fix: add studylocus to input loader in test

docs/python_api/datasets/l2g_feature.md

@@ -2,7 +2,27 @@
 title: L2G Feature
 ---
 
-::: gentropy.method.l2g.feature_factory.L2GFeature
+## Abstract Class
+
+::: gentropy.dataset.l2g_feature.L2GFeature
+
+## Feature Classes
+
+### Derived from colocalisation
+
+::: gentropy.dataset.l2g_feature.EQtlColocClppMaximumFeature
+::: gentropy.dataset.l2g_feature.PQtlColocClppMaximumFeature
+::: gentropy.dataset.l2g_feature.SQtlColocClppMaximumFeature
+::: gentropy.dataset.l2g_feature.TuQtlColocClppMaximumFeature
+::: gentropy.dataset.l2g_feature.EQtlColocH4MaximumFeature
+::: gentropy.dataset.l2g_feature.PQtlColocH4MaximumFeature
+::: gentropy.dataset.l2g_feature.SQtlColocH4MaximumFeature
+::: gentropy.dataset.l2g_feature.TuQtlColocH4MaximumFeature
+
+### Derived from distance
+
+::: gentropy.dataset.l2g_feature.DistanceTssMinimumFeature
+::: gentropy.dataset.l2g_feature.DistanceTssMeanFeature
 
 ## Schema
 

docs/python_api/methods/l2g/feature_factory.md

@@ -2,6 +2,6 @@
 title: L2G Feature Factory
 ---
 
-::: gentropy.method.l2g.feature_factory.ColocalisationFactory
+::: gentropy.method.l2g.feature_factory.FeatureFactory
 
-::: gentropy.method.l2g.feature_factory.StudyLocusFactory
+::: gentropy.method.l2g.feature_factory.L2GFeatureInputLoader

src/gentropy/config.py

@@ -227,50 +227,16 @@ class LocusToGeneConfig(StepConfig):
     gene_interactions_path: str | None = None
     features_list: list[str] = field(
         default_factory=lambda: [
-            # average distance of all tagging variants to gene TSS
-            "distanceTssMean",
-            # minimum distance of all tagging variants to gene TSS
-            "distanceTssMinimum",
-            # maximum vep consequence score of the locus 95% credible set among all genes in the vicinity
-            "vepMaximumNeighborhood",
-            # maximum vep consequence score of the locus 95% credible set split by gene
-            "vepMaximum",
-            # mean vep consequence score of the locus 95% credible set among all genes in the vicinity
-            "vepMeanNeighborhood",
-            # mean vep consequence score of the locus 95% credible set split by gene
-            "vepMean",
-            # max clpp for each (study, locus, gene) aggregating over all eQTLs
-            "eqtlColocClppMaximum",
-            # max clpp for each (study, locus) aggregating over all eQTLs
-            "eqtlColocClppMaximumNeighborhood",
-            # max clpp for each (study, locus, gene) aggregating over all pQTLs
-            "pqtlColocClppMaximum",
-            # max clpp for each (study, locus) aggregating over all pQTLs
-            "pqtlColocClppMaximumNeighborhood",
-            # max clpp for each (study, locus, gene) aggregating over all sQTLs
-            "sqtlColocClppMaximum",
-            # max clpp for each (study, locus) aggregating over all sQTLs
-            "sqtlColocClppMaximumNeighborhood",
-            # max clpp for each (study, locus) aggregating over all tuQTLs
-            "tuqtlColocClppMaximum",
-            # max clpp for each (study, locus, gene) aggregating over all tuQTLs
-            "tuqtlColocClppMaximumNeighborhood",
-            # max log-likelihood ratio value for each (study, locus, gene) aggregating over all eQTLs
-            "eqtlColocLlrMaximum",
-            # max log-likelihood ratio value for each (study, locus) aggregating over all eQTLs
-            "eqtlColocLlrMaximumNeighborhood",
-            # max log-likelihood ratio value for each (study, locus, gene) aggregating over all pQTLs
-            "pqtlColocLlrMaximum",
-            # max log-likelihood ratio value for each (study, locus) aggregating over all pQTLs
-            "pqtlColocLlrMaximumNeighborhood",
-            # max log-likelihood ratio value for each (study, locus, gene) aggregating over all sQTLs
-            "sqtlColocLlrMaximum",
-            # max log-likelihood ratio value for each (study, locus) aggregating over all sQTLs
-            "sqtlColocLlrMaximumNeighborhood",
-            # max log-likelihood ratio value for each (study, locus, gene) aggregating over all tuQTLs
-            "tuqtlColocLlrMaximum",
-            # max log-likelihood ratio value for each (study, locus) aggregating over all tuQTLs
-            "tuqtlColocLlrMaximumNeighborhood",
+            # max CLPP for each (study, locus, gene) aggregating over a specific qtl type
+            "eQtlColocClppMaximum",
+            "pQtlColocClppMaximum",
+            "sQtlColocClppMaximum",
+            "tuQtlColocClppMaximum",
+            # max H4 for each (study, locus, gene) aggregating over a specific qtl type
+            "eQtlColocH4Maximum",
+            "pQtlColocH4Maximum",
+            "sQtlColocH4Maximum",
+            "tuQtlColocH4Maximum",
         ]
     )
     hyperparameters: dict[str, Any] = field(
@@ -283,6 +249,7 @@ class LocusToGeneConfig(StepConfig):
     wandb_run_name: str | None = None
     hf_hub_repo_id: str | None = "opentargets/locus_to_gene"
     download_from_hub: bool = True
+    write_feature_matrix: bool = True
     _target_: str = "gentropy.l2g.LocusToGeneStep"