consIder X as a valid amino acid as per DeepGO-SE

- #36 (comment)

chebai/preprocessing/datasets/go_uniprot.py

@@ -1,13 +1,22 @@
-# Reference for this file :
+# References for this file :
+# Reference 1:
 # Maxat Kulmanov, Mohammed Asif Khan, Robert Hoehndorf;
 # DeepGO: Predicting protein functions from sequence and interactions
 # using a deep ontology-aware classifier, Bioinformatics, 2017.
 # https://doi.org/10.1093/bioinformatics/btx624
 # Github: https://github.com/bio-ontology-research-group/deepgo
+
+# Reference 2:
 # https://www.ebi.ac.uk/GOA/downloads
 # https://ftp.uniprot.org/pub/databases/uniprot/current_release/knowledgebase/complete/docs/keywlist.txt
 # https://www.uniprot.org/uniprotkb
 
+# Reference 3:
+# Kulmanov, M., Guzmán-Vega, F.J., Duek Roggli,
+# P. et al. Protein function prediction as approximate semantic entailment. Nat Mach Intell 6, 220–228 (2024).
+# https://doi.org/10.1038/s42256-024-00795-w
+# https://github.com/bio-ontology-research-group/deepgo2
+
 __all__ = [
     "GOUniProtOver250",
     "GOUniProtOver50",
@@ -34,6 +43,7 @@
 from chebai.preprocessing import reader as dr
 from chebai.preprocessing.datasets.base import _DynamicDataset
 
+# https://github.com/bio-ontology-research-group/deepgo/blob/master/utils.py#L15
 EXPERIMENTAL_EVIDENCE_CODES = {
     "EXP",
     "IDA",
@@ -43,6 +53,8 @@
     "IEP",
     "TAS",
     "IC",
+    # New evidence codes added in latest paper year 2024 Reference number 3
+    # https://github.com/bio-ontology-research-group/deepgo2/blob/main/deepgo/utils.py#L24-L26
     "HTP",
     "HDA",
     "HMP",
@@ -51,7 +63,9 @@
 }
 
 # https://github.com/bio-ontology-research-group/deepgo/blob/d97447a05c108127fee97982fd2c57929b2cf7eb/aaindex.py#L8
-AMBIGUOUS_AMINO_ACIDS = {"B", "O", "J", "U", "X", "Z", "*"}
+# https://github.com/bio-ontology-research-group/deepgo2/blob/main/deepgo/aminoacids.py#L10
+# `X` is now considered as valid amino acid, as per latest paper year 2024 Refernce number 3
+AMBIGUOUS_AMINO_ACIDS = {"B", "O", "J", "U", "Z", "*"}
 
 
 class _GOUniProtDataExtractor(_DynamicDataset, ABC):
@@ -416,12 +430,9 @@ def _get_swiss_to_go_mapping(self) -> pd.DataFrame:
 
         Note:
             This mapping is necessary because the GO data does not include the protein sequence representation.
-
-            Quote from the DeepGo Paper:
-            `We select proteins with annotations having experimental evidence codes
-            `EXPERIMENTAL_EVIDENCE_CODES` and filter the proteins by a
-            maximum length of 1002, ignoring proteins with ambiguous amino acid codes
-            (B, O, J, U, X, Z) in their sequence.`
+            We select proteins with annotations having experimental evidence codes, as specified in
+            `EXPERIMENTAL_EVIDENCE_CODES` and filter the proteins by a maximum length of 1002, ignoring proteins with
+            ambiguous amino acid codes specified in `AMBIGUOUS_AMINO_ACIDS` in their sequence.
 
             Check the link below for keyword details:
             https://ftp.uniprot.org/pub/databases/uniprot/current_release/knowledgebase/complete/docs/keywlist.txt

chebai/preprocessing/datasets/protein_pretraining.py

@@ -96,15 +96,15 @@ def _download_required_data(self) -> str:
     def _parse_protein_data_for_pretraining(self) -> pd.DataFrame:
         """
         Parses the Swiss-Prot data and returns a DataFrame containing Swiss-Prot proteins which does not have any valid
-        Gene Ontology(GO) label. A valid GO label is the one which has one of the following evidence code defined in
+        Gene Ontology(GO) label. A valid GO label is the one which has one of the following evidence codes, as specified in
         `EXPERIMENTAL_EVIDENCE_CODES`.
 
         The DataFrame includes the following columns:
             - "swiss_id": The unique identifier for each Swiss-Prot record.
             - "sequence": The protein sequence.
 
         Note:
-            We ignore proteins with ambiguous amino acid codes (B, O, J, U, X, Z) in their sequence.`
+            We ignore proteins with ambiguous amino acid specified in `AMBIGUOUS_AMINO_ACIDS` in their sequence.`
 
         Returns:
             pd.DataFrame: A DataFrame where each row corresponds to a Swiss-Prot record with not associated valid GO.

chebai/preprocessing/reader.py

@@ -348,7 +348,7 @@ class ProteinDataReader(DataReader):
 
     COLLATOR = RaggedCollator
 
-    # 20 natural amino acid notation
+    # 21 natural amino acid notation
     AA_LETTER = [
         "A",
         "R",
@@ -370,6 +370,8 @@ class ProteinDataReader(DataReader):
         "W",
         "Y",
         "V",
+        # https://github.com/bio-ontology-research-group/deepgo2/blob/main/deepgo/aminoacids.py#L3-L5
+        "X",  # Consider valid in latest paper year 2024 Reference number 3 in go_uniprot.py
     ]
 
     def name(self) -> str:

tests/unit/mock_data/ontology_mock_data.py

@@ -658,18 +658,19 @@ def get_UniProt_raw_data() -> str:
         - **Swiss_Prot_1**: A valid protein with three valid GO classes and one invalid GO class.
         - **Swiss_Prot_2**: Another valid protein with two valid GO classes and one invalid.
         - **Swiss_Prot_3**: Contains valid GO classes but has a sequence length > 1002.
-        - **Swiss_Prot_4**: Has valid GO classes but contains an invalid amino acid, 'X'.
+        - **Swiss_Prot_4**: Has valid GO classes but contains an invalid amino acid, 'B'.
         - **Swiss_Prot_5**: Has a sequence but no GO classes associated.
         - **Swiss_Prot_6**: Has GO classes without any associated evidence codes.
         - **Swiss_Prot_7**: Has a GO class with an invalid evidence code.
         - **Swiss_Prot_8**: Has a sequence length > 1002 and has only invalid GO class.
-        - **Swiss_Prot_9**: Has no GO classes but contains an invalid amino acid, 'X', in its sequence.
+        - **Swiss_Prot_9**: Has no GO classes but contains an invalid amino acid, 'B', in its sequence.
         - **Swiss_Prot_10**: Has a valid GO class but lacks a sequence.
         - **Swiss_Prot_11**: Has only Invalid GO class but lacks a sequence.
 
         Note:
-        A valid GO label is the one which has one of the following evidence code defined in
-        `EXPERIMENTAL_EVIDENCE_CODES`.
+        A valid GO label is the one which has one of the following evidence code specified in
+        go_uniprot.py->`EXPERIMENTAL_EVIDENCE_CODES`.
+        Invalid amino acids are specified in go_uniprot.py->`AMBIGUOUS_AMINO_ACIDS`.
 
         Returns:
             str: The raw UniProt data in string format.
@@ -715,7 +716,7 @@ def get_UniProt_raw_data() -> str:
             "DR   GO; GO:0000005; P:regulation of viral transcription; IEA:InterPro.\n"
             "DR   GO; GO:0000006; P:regulation of viral transcription; EXP:PomBase.\n"
             "SQ   SEQUENCE   60 AA;  29735 MW;  B4840739BF7D4121 CRC64;\n"
-            "     XAFSAEDVLK EYDRRRRMEA LLLSLYYPND RKLLDYKEWS PPRVQVECPK APVEWNNPPS\n"
+            "     BAFSAEDVLK EYDRRRRMEA LLLSLYYPND RKLLDYKEWS PPRVQVECPK APVEWNNPPS\n"
             "//\n"
             # Below protein with sequence string but has no GO class
             "ID   Swiss_Prot_5              Reviewed;         60 AA.\n"
@@ -749,7 +750,7 @@ def get_UniProt_raw_data() -> str:
             "ID   Swiss_Prot_9              Reviewed;         60 AA.\n"
             "AC   Q6GZX4;\n"
             "SQ   SEQUENCE   60 AA;  29735 MW;  B4840739BF7D4121 CRC64;\n"
-            "     XAFSAEDVLK EYDRRRRMEA LLLSLYYPND RKLLDYKEWS PPRVQVECPK APVEWNNPPS\n"
+            "     BAFSAEDVLK EYDRRRRMEA LLLSLYYPND RKLLDYKEWS PPRVQVECPK APVEWNNPPS\n"
             "//\n"
             # Below protein with a `valid` associated GO class but without sequence string
             "ID   Swiss_Prot_10              Reviewed;         60 AA.\n"