feat: protein stratification analysis for Davis#57

playground.py

@@ -1,21 +1,21 @@
-# # %%
-# from src.data_processing.init_dataset import create_datasets
+#%%
+from src.data_analysis.stratify_protein import check_davis_names, kinbase_to_df
 
-# create_datasets(
-#     data_opt=['davis'],
-#     feat_opt=['foldseek'],
-#     edge_opt=['binary']
-# )
-# # %%
-# from src.utils.loader import Loader
-# d2 = Loader.load_dataset(data='davis', pro_feature='nomsa',
-#                         edge_opt='anm')
+df = kinbase_to_df()
+# %%
+import json
+prot_dict = json.load(open('/home/jyaacoub/projects/data/davis/proteins.txt', 'r'))
 # %%
-from src.data_analysis.figures import fig4_pro_feat_violin, prepare_df
+# returns a dictionary of davis protein names (keys) and a truple of the protein name, main family, and subgroup (values)
+prots = check_davis_names(prot_dict, df)
+
+# %% plot histogram of main families and their counts
+import seaborn as sns
+import pandas as pd
+
+main_families = [v[1] for v in prots.values()]
+main_families = pd.Series(main_families)
+sns.histplot(main_families)
 
-df = prepare_df('results/model_media/model_stats.csv')
 
-#%%
-fig4_pro_feat_violin(df, sel_dataset='davis', sel_col='mse',
-                     add_stats=True, verbose=False)
 # %%

src/data_analysis/stratify_protein.py

@@ -0,0 +1,111 @@
+# To resolve https://github.com/jyaacoub/MutDTA/issues/57
+# list of protein names for davis is in ../data/davis/proteins.txt 
+# or downloaded from https://staff.cs.utu.fi/~aatapa/data/DrugTarget/target_gene_names.txt
+
+from typing import Iterable
+from src.utils import config as cfg
+import pandas as pd
+import re
+
+def kinbase_to_df(fasta_fp:str=f'{cfg.DATA_ROOT}/misc/Human_kinase_domain.fasta'):
+    """
+    Converts the KinBase fasta file containing all Human Kinase Domains into a 
+    dataframe for easy parsing.
+    
+    The Human_kinase_domain can be retrieved from:
+    https://web.archive.org/web/20230517032418/http://kinase.com/kinbase/FastaFiles/Human_kinase_domain.fasta
+        Sample of the FASTA headers:
+            >TTBK2_Hsap (CK1/TTBK) *
+            >TTBK1_Hsap (CK1/TTBK)
+            >TSSK4_Hsap (CAMK/TSSK)
+            >TSSK3_Hsap (CAMK/TSSK)
+        - The first couple characters before the underscore are the protein names (*e.g.: TTBK2).
+        - The characters after the underscore are the species (*e.g.: Hsap == homo sapiens).
+        - Most importantly, the characters between the parenthesis are the protein family and 
+          subgroups in that order (*e.g.: CK1/TTBK).
+
+    Parameters
+    ----------
+    `fasta_fp` : str, optional
+        The path to the downloaded fasta file path, by default f'{cfg.DATA_ROOT}/misc/Human_kinase_domain.fasta'
+    """
+    prots = {}
+    with open(fasta_fp, 'r') as f:
+        lines = f.readlines()
+        for i in range(len(lines)):
+            line = lines[i]
+            if line[0] == '>': # header
+                seq = lines[i+1]
+                name = re.search(r'^>(.+?)_Hsap', line).group(1)
+                # all in the fasta has a protein family discriptor with at least 2 elements
+                protein_family = re.search(r'\((.*)\)', line).group(1)
+                main_family, subgroup = protein_family.split('/')[:2]
+
+                prots[name] = (main_family, subgroup, seq)
+
+                i+=2
+            else:
+                i+=1
+    # convert to dataframe
+    df = pd.DataFrame.from_dict(prots, orient='index', columns=['main_family', 'subgroup', 'seq'])
+    df.index.name = 'protein_name'
+    return df
+
+def check_davis_names(davis_prots:dict, df:pd.DataFrame) -> list:
+    """
+    Checks davis protein names against fasta file containing human kinase domains. 
+    Returns list of proteins in davis that are also found in the fasta.
+    
+    NOTE: that for some in davis they have mutation information in brackets after 
+    the protein name, but we only need the protein name for this function.
+    Example:
+        ABL1(F317I)p -> ABL1
+    
+    There are also some with "-alpha?" (or "beta" etc..) where ? is a number, we 
+    can't ignore these and must include them by appending to the protein name "a?".
+        
+    Parameters
+    ----------
+    `davis_prots` : dict
+        Dictionary of davis protein names (keys) and their associated sequence (values).
+    `df` : pd.DataFrame
+        The dataframe containing the human kinase domains (see kinbase_to_df()).
+    """
+
+    df = kinbase_to_df() if df is None else df
+
+    greek = {'alpha', 'beta', 'gamma', 'delta'} # for checking if protein name has greek letter
+
+    found_prots = {}
+    for k in davis_prots.keys():
+        name = k.split('(')[0]
+        alpha = ''
+        if '-' in name:
+            name, alpha = name.split('-')
+
+        # removing any 'p' at the end of the name which indicates phosphorylation
+        if name[-1] == 'p':
+            name = name[:-1]
+
+        # getting alpha info if it exists
+        if len(alpha) >= 1:
+            alpha_name, alpha_num = re.search(r'([a-z]*)(\d*)', alpha).groups()
+            if alpha_name in greek:
+                name += f'{alpha_name[0]}{alpha_num}'
+
+        # checking if name is in the dataframe
+        if name in df.index:
+            found_prots[k] = (name, df.loc[name, 'main_family'], df.loc[name, 'subgroup'])
+        else:
+            print(f'MISSING: {k}-{name}')
+
+    return found_prots
+
+if __name__ == '__main__':
+    import json
+    from src.data_analysis.stratify_protein import check_davis_names, kinbase_to_df
+
+    prot_dict = json.load(open('/home/jyaacoub/projects/data/davis/proteins.txt', 'r'))
+
+    df = kinbase_to_df()
+    prots = check_davis_names(prot_dict, df)