Overlap between Davis, kiba and PDBbind

Maybe training on the combination of all 3 datasets will provide better results?

PDBbind, davis, kiba
Individual = 3643 379 228
Total concat = 4250

Total unique across all 3 = 4071
Total unique between pdbbind and davis = 4022
Total unique between pdbbind and kiba = 3871
Total unique between kiba and davis = 428

Shared between pdbbind and davis = 0
Shared between pdbbind and kiba = 0
Shared between kiba and davis = 179

Code:

# %% Checking overlap of proteins across davis, kiba and PDBbind
import pandas as pd

pdbbind_csv_p = "/cluster/home/t122995uhn/projects/data/PDBbindDataset/nomsa_binary_original_binary/full/XY.csv"
davis_csv_p =   "/cluster/home/t122995uhn/projects/data/DavisKibaDataset/davis/nomsa_binary_original_binary/full/XY.csv"
kiba_csv_p =    "/cluster/home/t122995uhn/projects/data/DavisKibaDataset/kiba/nomsa_binary_original_binary/full/XY.csv"

pdbbind_df = pd.read_csv(pdbbind_csv_p)
davis_df = pd.read_csv(davis_csv_p)
kiba_df = pd.read_csv(kiba_csv_p)

#%%
# dropping duplicates across prot_id and prot_seq independently
pdbbind_uni_id = pdbbind_df.prot_id.drop_duplicates(keep='first')
davis_uni_id = davis_df.prot_id.drop_duplicates(keep='first')
kiba_uni_id = kiba_df.prot_id.drop_duplicates(keep='first')

pdbbind_uni_seq = pdbbind_df.prot_seq.drop_duplicates(keep='first')
davis_uni_seq = davis_df.prot_seq.drop_duplicates(keep='first')
kiba_uni_seq = kiba_df.prot_seq.drop_duplicates(keep='first')

# %% keep only those indexes that show up in both unique id and unique seq.
pdbbind_uni = pd.merge(pdbbind_uni_id, pdbbind_uni_seq, left_index=True, right_index=True, how='inner')
davis_uni = pd.merge(davis_uni_id, davis_uni_seq, left_index=True, right_index=True, how='inner')
kiba_uni = pd.merge(kiba_uni_id, kiba_uni_seq, left_index=True, right_index=True, how='inner')

pdbbind_uni_seq = pdbbind_df.iloc[pdbbind_uni.index]['prot_seq']
davis_uni_seq = davis_df.iloc[davis_uni.index]['prot_seq']
kiba_uni_seq = kiba_df.iloc[kiba_uni.index]['prot_seq']

print("Individual =",len(pdbbind_uni_seq), len(davis_uni_seq), len(kiba_uni_seq))

#%%
all_uni_seq = pd.concat([pdbbind_uni_seq, davis_uni_seq, kiba_uni_seq])
all_uni_seq.reset_index(drop=True, inplace=True)

print("Total concat =", len(all_uni_seq))
print('Total unique across all 3 =', len(all_uni_seq.unique()))

# %%
unique_pdbbind_davis = pd.concat([pdbbind_uni_seq, davis_uni_seq]).unique()
unique_pdbbind_kiba = pd.concat([pdbbind_uni_seq, kiba_uni_seq]).unique()
unique_kiba_davis = pd.concat([kiba_uni_seq, davis_uni_seq]).unique()

print('Total unique between pdbbind and davis =', len(unique_pdbbind_davis))
print('Total unique between pdbbind and kiba =', len(unique_pdbbind_kiba))
print('Total unique between kiba and davis =', len(unique_kiba_davis))
print('')

print('Shared between pdbbind and davis =', (len(pdbbind_uni) + len(davis_uni)) - len(unique_pdbbind_davis))
print('Shared between pdbbind and kiba =', (len(pdbbind_uni) + len(kiba_uni)) - len(unique_pdbbind_kiba))
print('Shared between kiba and davis =', (len(kiba_uni) + len(davis_uni)) - len(unique_kiba_davis))

jyaacoub / MutDTA

Overlap between Davis, kiba and PDBbind #73

Code: