diff --git a/chebILP/ilp_problem_builder.py b/chebILP/ilp_problem_builder.py
index e89fe4c..3cff2b8 100644
--- a/chebILP/ilp_problem_builder.py
+++ b/chebILP/ilp_problem_builder.py
@@ -263,16 +263,16 @@ def gather_samples_for_chebi_cls(self, target_id: str, min_pos_samples=25, max_p
         sibling_neg_ids = set(sibling_neg_ids)
 
         samples_by_split = dict()
-        pos_train_samples = df_pos[df_pos.index.astype(str).isin(self.splits[self.splits["split"] == "train"])]
+        pos_train_samples = df_pos[df_pos.index.astype(str).isin(self.splits.loc[self.splits["split"] == "train", "id"])]
         samples_by_split[("pos", "train")] = pos_train_samples.sample(min(max_pos_samples, len(pos_train_samples)), random_state=42) # if there are more positives than max_pos_samples, sample randomly
-        neg_train_samples = df_neg[df_neg.index.astype(str).isin(self.splits[self.splits["split"] == "train"])]
+        neg_train_samples = df_neg[df_neg.index.astype(str).isin(self.splits.loc[self.splits["split"] == "train", "id"])]
         samples_by_split[("neg", "train")] = self.build_negative_mix(neg_train_samples, sibling_neg_ids, max_neg_samples)
         
-        samples_by_split[("pos", "validation")] = df_pos[df_pos.index.astype(str).isin(self.splits[self.splits["split"] == "validation"]) & df_pos.index.astype(str).isin(pos_ids)]
-        neg_val_samples = df_neg[df_neg.index.astype(str).isin(self.splits[self.splits["split"] == "validation"])]
+        samples_by_split[("pos", "validation")] = df_pos[df_pos.index.astype(str).isin(self.splits.loc[self.splits["split"] == "validation", "id"]) & df_pos.index.astype(str).isin(pos_ids)]
+        neg_val_samples = df_neg[df_neg.index.astype(str).isin(self.splits.loc[self.splits["split"] == "validation", "id"])]
         samples_by_split[("neg", "validation")] = self.build_negative_mix(neg_val_samples, sibling_neg_ids, max_neg_samples)
-        samples_by_split[("pos", "test")] = df_pos[df_pos.index.astype(str).isin(self.splits[self.splits["split"] == "test"]) & df_pos.index.astype(str).isin(pos_ids)]
-        neg_test_samples = df_neg[df_neg.index.astype(str).isin(self.splits[self.splits["split"] == "test"])]
+        samples_by_split[("pos", "test")] = df_pos[df_pos.index.astype(str).isin(self.splits.loc[self.splits["split"] == "test", "id"]) & df_pos.index.astype(str).isin(pos_ids)]
+        neg_test_samples = df_neg[df_neg.index.astype(str).isin(self.splits.loc[self.splits["split"] == "test", "id"])]
         samples_by_split[("neg", "test")] = self.build_negative_mix(neg_test_samples, sibling_neg_ids, max_neg_samples)
         
         for (posneg, split), df in samples_by_split.items():
diff --git a/chebILP/molecule_processing/data_preparation.py b/chebILP/molecule_processing/data_preparation.py
index 74dc963..e21ea01 100644
--- a/chebILP/molecule_processing/data_preparation.py
+++ b/chebILP/molecule_processing/data_preparation.py
@@ -190,7 +190,7 @@ def load_splits_from_csv(self) -> pd.DataFrame:
         if not os.path.exists(splits_path):
             raise FileNotFoundError(f"Splits file not found: {splits_path}. "
                                     f"Run `python -m chebILP prepare_dataset` to create it.")
-        splits_df = pd.read_csv(splits_path)
+        splits_df = pd.read_csv(splits_path, dtype={"id": str})
         if "id" not in splits_df.columns or "split" not in splits_df.columns:
             raise ValueError(f"Splits CSV must contain 'id' and 'split' columns: {splits_path}")
         if not all(s in splits_df["split"].unique() for s in ["train", "validation", "test"]):
diff --git a/pyproject.toml b/pyproject.toml
index 9c87406..4f4cea4 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -3,7 +3,7 @@ name = "chebilp"
 version = "1.1"
 description = "An Inductive Logic Programming framework for classifying chemical compounds into ChEBI classes."
 readme = "README.md"
-requires-python = ">=3.10"
+requires-python = ">=3.11"
 dependencies = [
     "chebi-utils>=0.2.1",
     "clingo>=5.8.0",