improve melting point templates

OpenBioML · Aug 13, 2024 · db352d4 · db352d4
1 parent a21a01d
commit db352d4
Show file tree

Hide file tree

Showing 4 changed files with 131 additions and 48 deletions.
diff --git a/data/tabular/melting_points/meta.yaml b/data/tabular/melting_points/meta.yaml
@@ -7,7 +7,7 @@ targets:
     units: deg C
     type: continuous
     names:
-      - noun: mean melting point
+      - noun: melting point
     uris:
   - id: mp_range
     description: melting point range
@@ -45,45 +45,69 @@ bibtex:
     }
 templates:
   - |-
-    {#Task: |Task: |!}{#Predict|Estimate!} the melting point of {NAME#}.
-    {#Answer: |A: |!}The melting point is {mp#} deg C.
+    {#Task: |Task: |!}{#Predict|Estimate!} the {mp__names__noun} of {NAME#}.
+    {#Answer: |A: |!}The {mp__names__noun} is {mp#} {mp__units}.
   - |-
-    {#Task: |Task: |!}{#Predict|Estimate!} the melting point of a {#molecule|compound!} with the {SMILES__description} {SMILES#}?
-    {#Answer: |A: |!}{#The melting point is |!}{mp#} deg C.
+    {#Task: |Task: |!}{#Predict|Estimate!} the {mp__names__noun} of a {#molecule|compound!} with the {SMILES__description} {SMILES#}?
+    {#Answer: |A: |!}{#The {mp__names__noun} is |!}{mp#} {mp__units}.
   - |-
-    {#Question: |Q: !}What is the melting point of {NAME#}?
-    {#Answer: |A: |!}{#The melting point is |!}{mp#} deg C.
+    {#Question: |Q: !}What is the {mp__names__noun} of {NAME#}?
+    {#Answer: |A: |!}{#The melting point is |!}{mp#} {mp__units}.
   - |-
-    {#Question: |Q: !}What is the melting point of a {#molecule|compound!} with the {SMILES__description} {SMILES#}?
-    {#Answer: |A: |!}{#The melting point is |!}{mp#} deg C.
+    {#Question: |Q: !}What is the {mp__names__noun} of a {#molecule|compound!} with the {SMILES__description} {SMILES#}?
+    {#Answer: |A: |!}{#The melting point is |!}{mp#} {mp__units}.
   - |-
-    {#Question: |Q: !}What is the melting point of {NAME#}?
-    {#Answer: |A: |!}{#The melting point is in the range |!}{mp_range#} deg C.
+    {#Question: |Q: !}What is the {mp__names__noun} of {NAME#}?
+    {#Answer: |A: |!}{#The melting point is in the range |!}{mp_range#} {mp__units}.
   - |-
-    {#Question: |Q: !}What is the melting point of a {#molecule|compound!} with the {SMILES__description} {SMILES#}?
-    {#Answer: |A: |!}{#The melting point is in the range |!}{mp_range#} deg C.
+    {#Question: |Q: !}What is the {mp__names__noun} of a {#molecule|compound!} with the {SMILES__description} {SMILES#}?
+    {#Answer: |A: |!}{#The melting point is in the range |!}{mp_range#} {mp__units}.
   - |-
-    {#Question: |Q: !}What is a compound with a melting point of {mp#} deg C?
+    {#Question: |Q: !}What is a compound with a {mp__names__noun} of {mp#} {mp__units}?
     {#Answer: |A: |!}{NAME#}
   - |-
-    {#Question: |Q: !}What is a compound with a melting point in the range {mp_range#} deg C?
+    {#Question: |Q: !}What is a compound with a {mp__names__noun} in the range {mp_range#} {mp__units}?
     {#Answer: |A: |!}{NAME#}
   - |-
     User: I have a question about {NAME#}.
     Assistant: {#Sure, what is your question?|How can I help?|That sounds interesting, how can I help?|Interesting, how can I help?!}
-    User: What is the melting point of {#this compound|this molecule!}?
-    Assistant: {#The melting point is |!}{mp#} deg C.
+    User: What is the {mp__names__noun} of {#this compound|this molecule!}?
+    Assistant: {#The melting point is |!}{mp#} {mp__units}.
   - |-
     User: I have a question about a {#compound|molecule!} with the {SMILES__description} {SMILES#}.
     Assistant: {#Sure, what is your question?|How can I help?|That sounds interesting, how can I help?|Interesting, how can I help?!}
-    User: What is the melting point of {#this compound|this molecule!}?
-    Assistant: {#The melting point is |!}{mp#} deg C. {#Is there anything else I can help you with?|Do you have any other questions?|Do you have any other questions for me?|Is there anything else I can help you with today?|Do you have any other questions for me today?!}
+    User: What is the {mp__names__noun} of {#this compound|this molecule!}?
+    Assistant: {#The melting point is |!}{mp#} {mp__units}. {#Is there anything else I can help you with?|Do you have any other questions?|Do you have any other questions for me?|Is there anything else I can help you with today?|Do you have any other questions for me today?!}
     User: {#Yes,|Indeed,!} what is the name of {#this compound|this molecule!}?
     Assistant: {NAME#}
   - |-
     User: I have a question about {NAME#}.
     Assistant: {#Sure, what is your question?|How can I help?|That sounds interesting, how can I help?|Interesting, how can I help?!}
-    User: What is the melting point of {#this compound|this molecule!}?
-    Assistant: {#The melting point is |!}{mp_range#} deg C. {#Is there anything else I can help you with?|Do you have any other questions?|Do you have any other questions for me?|Is there anything else I can help you with today?|Do you have any other questions for me today?!}
+    User: What is the {mp__names__noun} of {#this compound|this molecule!}?
+    Assistant: {#The melting point is |!}{mp_range#} {mp__units}. {#Is there anything else I can help you with?|Do you have any other questions?|Do you have any other questions for me?|Is there anything else I can help you with today?|Do you have any other questions for me today?!}
     User: {#Yes,|Indeed,!} what is the {SMILES__description} of {#this compound|this molecule!}?
     Assistant: {SMILES#}
+  - |-
+    Task: Please estimate the {mp_names__noun} of a compound.
+
+    Compound: {NAME#}
+
+    Result:<EOI> {mp#} {mp__units}
+  - |-
+    Task: Please estimate the {mp_names__noun} of a compound.
+
+    {SMILES__description}: {SMILES#}
+
+    Result:<EOI> {mp#} {mp__units}
+  - |-
+    Question: What is the {mp_names__noun} of a compound with the {SMILES__description} {SMILES#} in {mp__units}?
+
+    Answer:<EOI> {mp#}
+  - |-
+    Question: Which molecule has a {mp_names__noun} of {mp#} {mp__units}?
+    Pick {%multiple_choice_enum%3%aA1}.
+
+    Options:
+    {SMILES%}
+
+    Answer: {%multiple_choice_result}
diff --git a/src/chemnlp/data/meta_yaml_augmentor.py b/src/chemnlp/data/meta_yaml_augmentor.py
@@ -76,6 +76,8 @@
       This is a multi-line string.
       It can contain multiple lines.
 
+Stick to the syntax of the templating language that is described above. Do not use any other syntax and do not remove the special characters.
+
 Just return raw YAML string, do not wrap it into backticks or anything else.
 """
 

diff --git a/src/chemnlp/data/sampler.py b/src/chemnlp/data/sampler.py
@@ -71,10 +71,22 @@ def _get_additional_targets(self, df: pd.DataFrame) -> List[str]:
     def _add_additional_targets_to_meta(self):
         additional_targets_meta = {
             "selfies": {"id": "selfies", "type": "selfies", "description": "SELFIES"},
-            "deepsmiles": {"id": "deepsmiles", "type": "deepsmiles", "description": "DeepSMILES"},
-            "canonical": {"id": "canonical", "type": "canonical", "description": "canonical SMILES"},
+            "deepsmiles": {
+                "id": "deepsmiles",
+                "type": "deepsmiles",
+                "description": "DeepSMILES",
+            },
+            "canonical": {
+                "id": "canonical",
+                "type": "canonical",
+                "description": "canonical SMILES",
+            },
             "inchi": {"id": "inchi", "type": "inchi", "description": "InChI"},
-            "iupac_name": {"id": "iupac_name", "type": "iupac_name", "description": "IUPAC name"},
+            "iupac_name": {
+                "id": "iupac_name",
+                "type": "iupac_name",
+                "description": "IUPAC name",
+            },
         }
         for target in self.additional_targets:
             self.meta["targets"].append(additional_targets_meta[target])
@@ -538,7 +550,11 @@ def sample(self, sample: pd.Series, template: str) -> str:
         if sample is None:
             sample = self.df.sample(1).iloc[0]
         if self.additional_targets and "SMILES" in sample.index:
-            non_nan_targets = [target for target in ["SMILES"] + self.additional_targets if pd.notna(sample[target])]
+            non_nan_targets = [
+                target
+                for target in ["SMILES"] + self.additional_targets
+                if pd.notna(sample[target])
+            ]
             new_target = random.choice(non_nan_targets)
             if new_target != "SMILES":
                 template = template.replace("{SMILES", "{" + new_target)

diff --git a/tests/data/test_sampler.py b/tests/data/test_sampler.py
@@ -19,17 +19,30 @@ def sample_df():
         }
     )
 
+
 @pytest.fixture
 def sample_multiple_identifier_df():
-    return pd.DataFrame({
-        'SMILES': ['CC(C)NCC(O)c1ccc(O)c(O)c1', 'CC1=C(C(=O)NC2=C1C=CC=C2)C3=CC=CC=C3'],
-        'selfies': ['[C][C][Branch1][C][C][N][C][C][Branch1][O][C][=C][C][=C][C][Branch1][O][C][=C][Branch1][O][C][=C]', '[C][C]1[=C][Branch1][C][Branch2][=O][N][C]2[=C]1[C][=C][C][=C][C]2[C]3[=C][C][=C][C][=C][C]3'],
-        'inchi': ['InChI=1S/C11H17NO3/c1-8(2)12-6-9(13)10-4-3-5-11(14)7-10/h3-5,7-9,12-14H,6H2,1-2H3', 'InChI=1S/C15H11NO/c17-14-11-9-5-1-3-7-13(9)16-15(14)10-6-2-4-8-12(10)11/h1-8,16H'],
-        'compound_name': ['Isoproterenol', 'Phenytoin'],
-        'LogP': [0.08, 2.47],
-        'is_active': [True, False],
-        'split': ['train', 'test']
-    })
+    return pd.DataFrame(
+        {
+            "SMILES": [
+                "CC(C)NCC(O)c1ccc(O)c(O)c1",
+                "CC1=C(C(=O)NC2=C1C=CC=C2)C3=CC=CC=C3",
+            ],
+            "selfies": [
+                "[C][C][Branch1][C][C][N][C][C][Branch1][O][C][=C][C][=C][C][Branch1][O][C][=C][Branch1][O][C][=C]",
+                "[C][C]1[=C][Branch1][C][Branch2][=O][N][C]2[=C]1[C][=C][C][=C][C]2[C]3[=C][C][=C][C][=C][C]3",
+            ],
+            "inchi": [
+                "InChI=1S/C11H17NO3/c1-8(2)12-6-9(13)10-4-3-5-11(14)7-10/h3-5,7-9,12-14H,6H2,1-2H3",
+                "InChI=1S/C15H11NO/c17-14-11-9-5-1-3-7-13(9)16-15(14)10-6-2-4-8-12(10)11/h1-8,16H",
+            ],
+            "compound_name": ["Isoproterenol", "Phenytoin"],
+            "LogP": [0.08, 2.47],
+            "is_active": [True, False],
+            "split": ["train", "test"],
+        }
+    )
+
 
 @pytest.fixture
 def sample_meta():
@@ -69,19 +82,28 @@ def sample_meta():
 def sample_multiple_identifier_meta():
     return {
         "targets": [
-            {"id": "LogP", "type": "continuous", "description": "Logarithm of partition coefficient"},
-            {"id": "is_active", "type": "categorical", "description": "Activity status"}
+            {
+                "id": "LogP",
+                "type": "continuous",
+                "description": "Logarithm of partition coefficient",
+            },
+            {
+                "id": "is_active",
+                "type": "categorical",
+                "description": "Activity status",
+            },
         ],
         "identifiers": [
             {"id": "SMILES", "type": "SMILES", "description": "SMILES notation"},
-            {"id": "compound_name", "type": "Other", "description": "Compound name"}
+            {"id": "compound_name", "type": "Other", "description": "Compound name"},
         ],
         "templates": [
             "The molecule with SMILES {SMILES#} has a LogP of {LogP#}.",
-            "The compound {compound_name#} is {is_active#active&inactive}."
-        ]
+            "The compound {compound_name#} is {is_active#active&inactive}.",
+        ],
     }
 
+
 @pytest.fixture
 def sample_config():
     return {
@@ -504,21 +526,40 @@ def test_polymer_multiple_properties(
     assert "275.0" in result
     assert "0.90" in result
 
-def test_additional_targets(sample_multiple_identifier_df, sample_multiple_identifier_meta, sample_config):
-    sampler = TemplateSampler(sample_multiple_identifier_df, sample_multiple_identifier_meta, sample_config)
+
+def test_additional_targets(
+    sample_multiple_identifier_df, sample_multiple_identifier_meta, sample_config
+):
+    sampler = TemplateSampler(
+        sample_multiple_identifier_df, sample_multiple_identifier_meta, sample_config
+    )
     assert set(sampler.additional_targets) == {"selfies", "inchi"}
-    print(sampler.meta['targets'])
+    print(sampler.meta["targets"])
     assert len(sampler.meta["targets"]) == 4
 
 
-
-def test_sample_with_random_replacement(sample_multiple_identifier_df, sample_multiple_identifier_meta, sample_config):
-    sampler = TemplateSampler(sample_multiple_identifier_df, sample_multiple_identifier_meta, sample_config)
+def test_sample_with_random_replacement(
+    sample_multiple_identifier_df, sample_multiple_identifier_meta, sample_config
+):
+    sampler = TemplateSampler(
+        sample_multiple_identifier_df, sample_multiple_identifier_meta, sample_config
+    )
     template = "The compound with {SMILES__description} {SMILES#} has a {LogP__description} of {LogP#}"
-    results = [sampler.sample(sample_multiple_identifier_df.iloc[0], template) for _ in range(20)]
+    results = [
+        sampler.sample(sample_multiple_identifier_df.iloc[0], template)
+        for _ in range(20)
+    ]
     smiles_count = sum("CC(C)NCC(O)c1ccc(O)c(O)c1" in r for r in results)
-    selfies_count = sum("[C][C][Branch1][C][C][N][C][C][Branch1][O][C][=C][C][=C][C][Branch1][O][C][=C][Branch1][O][C][=C]" in r for r in results)
-    inchi_count = sum("InChI=1S/C11H17NO3/c1-8(2)12-6-9(13)10-4-3-5-11(14)7-10/h3-5,7-9,12-14H,6H2,1-2H3" in r for r in results)
+    selfies_count = sum(
+        "[C][C][Branch1][C][C][N][C][C][Branch1][O][C][=C][C][=C][C][Branch1][O][C][=C][Branch1][O][C][=C]"
+        in r
+        for r in results
+    )
+    inchi_count = sum(
+        "InChI=1S/C11H17NO3/c1-8(2)12-6-9(13)10-4-3-5-11(14)7-10/h3-5,7-9,12-14H,6H2,1-2H3"
+        in r
+        for r in results
+    )
     assert smiles_count > 0
     assert selfies_count > 0
     assert inchi_count > 0