Merge pull request #5 from matchms/backup

backup because of laptop crash

Author: niekdejonge

ms2query/benchmarking/Embeddings.py

@@ -31,6 +31,7 @@ def combine_embeddings(cls, embeddings_1: "Embeddings", embeddings_2: "Embedding
         if embeddings_1.model_settings != embeddings_2.model_settings:
             raise ValueError("Model settings of merged embeddings do not match")
         if not set(embeddings_1.index_to_spectrum_hash).isdisjoint(embeddings_2.index_to_spectrum_hash):
+            # todo allow this to happen, but remove repeating ones and check that they are the same.
             raise ValueError("There are repeated spectra in the embeddings that are added together")
         combined_embeddings =  np.vstack([embeddings_1.embeddings, embeddings_2.embeddings])
         index_to_spectrum_hash = embeddings_1.index_to_spectrum_hash + embeddings_2.index_to_spectrum_hash

ms2query/benchmarking/reference_methods/predict_with_integrated_similarity_flow.py

@@ -2,13 +2,16 @@
 import numpy as np
 from matchms.similarity.vector_similarity_functions import jaccard_similarity_matrix
 from tqdm import tqdm
+
+from ms2query.benchmarking.Fingerprints import Fingerprints
 from ms2query.benchmarking.reference_methods.PredictMS2DeepScoreSimilarity import predict_top_ms2deepscores
 from ms2query.benchmarking.AnnotatedSpectrumSet import AnnotatedSpectrumSet
 
 
 def predict_with_integrated_similarity_flow(
     library_spectra: AnnotatedSpectrumSet,
     query_spectra: AnnotatedSpectrumSet,
+    fingerprints: Fingerprints,
     number_of_analogues_to_consider=50,
 ) -> Tuple[List[str], List[float]]:
 
@@ -22,6 +25,7 @@ def predict_with_integrated_similarity_flow(
         highest_isf_score, inchikey_of_highest_isf_score = get_highest_isf(
             library_spectra,
             all_indexes_of_library_spectra_with_highest_score[query_index],
+            fingerprints,
             all_predicted_scores[query_index],
         )
         inchikeys_of_best_matches.append(inchikey_of_highest_isf_score)
@@ -32,7 +36,8 @@ def predict_with_integrated_similarity_flow(
 def get_highest_isf(
     library_spectra: AnnotatedSpectrumSet,
     indexes_of_library_spectra_with_highest_score: np.ndarray,
-    predicted_scores: [List[float]],
+        fingerprints: Fingerprints,
+        predicted_scores: [List[float]],
 ):
 
     # Get the corresponding inchikeys
@@ -43,7 +48,8 @@ def get_highest_isf(
         predicted_scores, inchikeys_with_highest_ms2deepscore
     )
     # calculate tanimoto scores
-    tanimoto_scores = jaccard_similarity_matrix(library_spectra.fingerprints.fingerprints, library_spectra.fingerprints.fingerprints)
+    library_fingerprints = fingerprints.get_fingerprints(library_spectra.inchikeys)
+    tanimoto_scores = jaccard_similarity_matrix(library_fingerprints, library_fingerprints)
 
     isf_scores = integrated_similarity_flow(average_scores, tanimoto_scores, nr_of_spectra_per_inchikey)
     index_of_highest_score = np.argmax(isf_scores)

tests/test_methods.py

@@ -1,6 +1,8 @@
 import numpy as np
 import pytest
 from matchms.similarity.vector_similarity_functions import jaccard_similarity_matrix
+
+from ms2query.benchmarking.Fingerprints import Fingerprints
 from ms2query.benchmarking.reference_methods.predict_best_possible_match import predict_best_possible_match
 from ms2query.benchmarking.reference_methods.predict_highest_cosine import predict_highest_cosine
 from ms2query.benchmarking.reference_methods.predict_highest_ms2deepscore import predict_highest_ms2deepscore
@@ -11,36 +13,43 @@
 from ms2query.benchmarking.AnnotatedSpectrumSet import AnnotatedSpectrumSet
 from tests.conftest import create_test_spectra, ms2deepscore_model
 
+def get_library_and_test_spectra():
+    model = ms2deepscore_model()
+    library_spectra = AnnotatedSpectrumSet.create_spectrum_set(create_test_spectra())
+    test_spectra = AnnotatedSpectrumSet.create_spectrum_set(create_test_spectra(1))
+    library_spectra.add_embeddings(model)
+    test_spectra.add_embeddings(model)
+    return library_spectra, test_spectra
 
 @pytest.mark.parametrize(
     "prediction_function",
     [
         predict_highest_cosine,
         predict_highest_ms2deepscore,
-        predict_best_possible_match,
     ],
 )
 def test_all_methods(prediction_function):
-    model = ms2deepscore_model()
-
-    library_spectra = AnnotatedSpectrumSet.create_spectrum_set(create_test_spectra())
-    test_spectra = AnnotatedSpectrumSet.create_spectrum_set(create_test_spectra(1))
-    library_spectra.add_embeddings(model)
-    test_spectra.add_embeddings(model)
+    library_spectra, test_spectra = get_library_and_test_spectra()
     predicted_inchikeys, scores = prediction_function(library_spectra, test_spectra)
     for i, spectrum in enumerate(test_spectra.spectra):
         inchikey = spectrum.get("inchikey")[:14]
         assert predicted_inchikeys[i] == inchikey
         assert np.allclose(scores[i], np.array(1.0), atol=1e-5)
 
 
+def test_predict_best_possible_match():
+    library_spectra, test_spectra = get_library_and_test_spectra()
+    fingerprints = Fingerprints.from_spectrum_set(library_spectra + test_spectra, "daylight", 2048)
+    predicted_inchikeys, scores = predict_best_possible_match(library_spectra, test_spectra, fingerprints)
+    for i, spectrum in enumerate(test_spectra.spectra):
+        inchikey = spectrum.get("inchikey")[:14]
+        assert predicted_inchikeys[i] == inchikey
+        assert np.allclose(scores[i], np.array(1.0), atol=1e-5)
+
 def test_predict_with_integrated_similarity_flow():
-    model = ms2deepscore_model()
-    library_spectra = AnnotatedSpectrumSet.create_spectrum_set(create_test_spectra())
-    test_spectra = AnnotatedSpectrumSet.create_spectrum_set(create_test_spectra(1))
-    library_spectra.add_embeddings(model)
-    test_spectra.add_embeddings(model)
-    predicted_inchikeys, scores = predict_with_integrated_similarity_flow(library_spectra, test_spectra)
+    library_spectra, test_spectra = get_library_and_test_spectra()
+    fingerprints = Fingerprints.from_spectrum_set(library_spectra, "daylight", 4096)
+    predicted_inchikeys, scores = predict_with_integrated_similarity_flow(library_spectra, test_spectra, fingerprints)
 
     assert predicted_inchikeys == ["RYYVLZVUVIJVGH", "ZPUCINDJVBIVPJ", "ZPUCINDJVBIVPJ"]
     assert np.allclose(np.array([0.38829751082577607, 0.3919729335980483, 0.38774130710967564]), np.array(scores))

tests/test_predict_using_closest_tanimoto.py

@@ -1,5 +1,7 @@
 import numpy as np
 import pytest
+
+from ms2query.benchmarking.Fingerprints import Fingerprints
 from ms2query.benchmarking.reference_methods.predict_using_closest_tanimoto import (
     get_average_predictions_for_closely_related_metabolites,
     get_inchikey_and_tanimoto_scores_for_top_k,
@@ -18,7 +20,8 @@ def test_predict_using_closest_tanimoto():
     test_spectra = AnnotatedSpectrumSet.create_spectrum_set(create_test_spectra(1, nr_of_inchikeys=3))
     library_spectra.add_embeddings(model)
     test_spectra.add_embeddings(model)
-    predicted_inchikeys, scores = predict_using_closest_tanimoto(library_spectra, test_spectra, 3, 3)
+    fingerprints = Fingerprints.from_spectrum_set(library_spectra, "daylight", 2048)
+    predicted_inchikeys, scores = predict_using_closest_tanimoto(library_spectra, test_spectra, fingerprints, 3, 3)
 
     assert isinstance(predicted_inchikeys, list)
     assert len(predicted_inchikeys) == 3
@@ -32,7 +35,9 @@ def test_predict_using_closest_tanimoto_single_spectrum():
     test_spectra = AnnotatedSpectrumSet.create_spectrum_set(create_test_spectra(1, nr_of_inchikeys=3))
     library_spectra.add_embeddings(model)
     test_spectra.add_embeddings(model)
-    predicted_inchikey, score = predict_using_closest_tanimoto_single_spectrum(library_spectra, test_spectra, 3, 3)
+    fingerprints = Fingerprints.from_spectrum_set(library_spectra, "daylight", 2048)
+
+    predicted_inchikey, score = predict_using_closest_tanimoto_single_spectrum(library_spectra, test_spectra, 3, 3, fingerprints)
 
     assert isinstance(predicted_inchikey, str)
     assert len(predicted_inchikey) ==14
@@ -57,7 +62,7 @@ def test_get_average_predictions_for_closely_related_metabolites():
     test_spectra = test_spectra.copy()[2:]
     spectra = AnnotatedSpectrumSet.create_spectrum_set(test_spectra)
 
-    inchikeys = list(spectra.inchikey_fingerprint_pairs.keys())[:3]
+    inchikeys = spectra.inchikeys[:3]
     ms2deepscores = np.zeros(len(spectra.spectra))
     ms2deepscores[0] = 0.8
     ms2deepscores[[1,2,3]] = 0.6