Upload jolma_domains.py

jolma_domains.py

@@ -0,0 +1,73 @@
+import pandas as pd
+import datasets
+
+logger = datasets.logging.get_logger(__name__)
+
+URL = "https://huggingface.co/datasets/thewall/jolma_domains/resolve/main"
+Protein_URL = "https://huggingface.co/datasets/thewall/DeepBindWeight/resolve/main"
+
+
+class JolmaDomainsConfig(datasets.BuilderConfig):
+    pass
+
+
+class JolmaDomains(datasets.GeneratorBasedBuilder):
+    BUILDER_CONFIGS = [
+        JolmaDomainsConfig(name=key) for key in ["Region", "Compositional bias", "Zinc finger", "Domain", "Motif", "Repeat", "Coiled coil"]
+    ]
+
+    DEFAULT_CONFIG_NAME = "Region"
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            features=datasets.Features(
+                {
+                    "id": datasets.Value("int32"),
+                    "protein": datasets.Value("string"),
+                    "protein_id": datasets.Value("string"),
+                    "type": datasets.Value("string"),
+                    "detail": datasets.Sequence({
+                        "start": datasets.Value("int32"),
+                        "end": datasets.Value("int32"),
+                        "span": datasets.Value("string"),
+                        "description": datasets.Value("string"),
+                        "evidence": datasets.Value("string")}),
+                }
+            ),
+        )
+
+    def _split_generators(self, dl_manager):
+        file = dl_manager.download(f"{URL}/domains.xlsx")
+        protein_file = dl_manager.download(f"{Protein_URL}/ERP001824-UniprotKB.xlsx")
+        return [
+            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": file,"protein_filepath": protein_file}),
+        ]
+
+    def _generate_examples(self, filepath, protein_filepath):
+        """This function returns the examples in the raw (text) form."""
+        protein_info = pd.read_excel(protein_filepath, index_col=0)
+        data = pd.read_excel(filepath, sheet_name=self.config.name)
+
+        groups = data.groupby("Entry")
+        for idx, (entry, rows) in enumerate(groups):
+            protein = protein_info[protein_info['Entry'] == entry].iloc[0]
+            sequence = protein['Sequence']
+            detail = []
+            for _, row in rows.iterrows():
+                detail.append({"span": row["span"],
+                               "description": row['description'],
+                               "evidence": row['evidence']})
+                span = list(map(int, row["span"].split("-")))
+                detail[-1]['start'] = span[0]-1
+                detail[-1]['end'] = span[1]
+            yield idx, {"id": idx,
+                        "protein": sequence,
+                        "protein_id": entry,
+                        "type": self.config.name,
+                        "detail": detail
+                        }
+
+
+if __name__=="__main__":
+    from datasets import load_dataset
+    dataset = load_dataset("jolma_domains.py", split="all")