License: CC BY-NC-ND 4.0
bbbPythoN is a Python package for the prediction of Blood-Brain barrier (BBB) permeation of small molecules.
Based on fingerprints encoding values of a pre-selected set of descriptors chosen for their discriminatory
power in regard to BBB permeability, random forest classifiers predict whether compounds are likely to
cross the BBB. Two different models, trained on an imbalanced and a balanced dataset, respectively, are
included in this package.
As it was shown in "Non-Animal Models for Blood-Brain Barrier Permeability Evaluation of
Drug-Like Compounds" our models are most suited for the prediction of passive permeation.
Given a set of compounds, either as SMILES via command line input, as .csv file containing SMILES, or as
.sdf file containing MolBlocks, fingerprints can be produced and the molecules' activity predicted via
the random forests.
After downloading the bbbPackage-1.0-py3.whl file from the dist folder of the repository, the package can be
installed via the command pip install filepath/bbbPackage-1.0-py3-none-any.whl. Where filepath corresponds to the path
to the folder the .whl file is located in.
Additional packages used in this package will be installed as well and consist of:
joblib, numpy, pandas, openpyxl, rdkit, mordred
This package consists of the functionalities necessary to create the Blood-Brain barrier (bbb) MI-DSE
Fingerprints described in ""Non-Animal Models for Blood-Brain Barrier Permeability Evaluation of
Drug-Like Compounds" for a given SMILES, set of SMILES or set of molecules supplied via .sdf and predict
their bbb-permeability.
Function ProdFP() produces the aforementioned fingerprints and accepts either a single SMILES string, a .csv
file containing SMILES strings and their corresponding IDs, or a .sdf file containing molecules and their
corresponding IDs. In both the latter cases activities can be supplied as well.
ProdFP() returns a pandas Dataframe holding the descriptors the fingerprints are based on, the fingerprints
as well as the corresponding IDs and activities.
ProdFP() has the following keyword parameters:
smiles:
A single input SMILES string, if a set containing multiple molecules should be fingerprinted use filepath.
By default an empty string.
filepath:
Filepath specifying location of the .csv or .sdf file used as input. By default an empty string.
id_name:
Either specifies the molecule property holding the IDs in the .sdf file.
Or the name of the column in the .csv file containing the IDs.
By default an empty string.
act_name:
Either specifies the molecule property holding the activities in the .sdf file.
Or the name of the column in the .csv file containing the activities.
By default an empty string.
smiles_name:
Only needed in case a .csv file is used as input. Specifies the name of the
column containing the SMILES strings in the .csv file.
use_bal:
Specifies whether to use the boundaries for binning of descriptor values based
on the balanced, or imabalanced dataset.
By default False.
The function BbbPred() predicts bbb-permeability based on the fingerprints contained in the output Dataframe
of ProdFP(). Its parameters are:
inp_df:
Corresponds to the Dataframe returned by ProdFP().
act:
If set to True, BbbPred() will produce a .csv, and a .xlsx file detailing the performance of the model.
Only in case an input file containing activities was supplied can this option be used. By default False.
ret:
If set to True, BbbPred() returns the IDs, predictions, and prediction probabilities as lists.
Otherwise each molecule's ID, the corresponding prediction and prediction probability are printed.
By default False.
use_bal:
Specifies whether to use the model trained on the balanced, or imbalanced dataset.
By default False.
import os
from bbbPythoN import bbbPythoN
# Defining filepath of input .csv.
filepath_csv = os.path.join("dir1","input.csv")
# Defining filepath of input .sdf.
filepath_sdf = os.path.join("dir2","input.sdf")
# As mordred uses parallelization in its descriptor calculation the "if __name__ == '__main__':" statement
# is necessary to not spawn subprocesses from outside the main script.
if __name__ == '__main__':
# Calculating fingerprints and specifying column names of SMILES, ID, and activity in .csv.
# In case no activities are available omit act_name.
# Exchange "id_name", "act_name", and "smiles_name" with the respective column names
# of your dataset.
csv_fp_df = bbbPythoN.ProdFP(filepath=filepath_csv,id_name="id_name",act_name="act_name",smiles_name="act_name",use_bal=True)
# Calculating fingerprints and molecule properties for compounds contained in .sdf file.
# In case no activities are available omit act_name.
# Exchange "id_name", "act_name", and "smiles_name" with the respective column names
# of your dataset.
sdf_fp_df = bbbPythoN.ProdFP(filepath=filepath_sdf,id_name="id_name",act_name="act_name",smiles_name="act_name",use_bal=True)
# It is advised to save the pandas DataFrames using .to_pickle() instead of .to_csv(), as the latter
# will cast the python list objects representing the Fingerprints to python strings.
csv_fp_df.to_pickle(os.path.join("dir1","output.pkl"))
sdf_fp_df.to_pickle(os.path.join("dir2","output.pkl"))
# Predicting bbb-permeability of .csv compounds.
# If act=True, a .csv file listing True Positives, True Negatives, False Positives, and
# False Negatives and a .xlsx containing performance metrics are produced.
# Setting ret=True, returns the IDs, predictions (result), and prediction probabilities (probas).
ids, result, probas = bbbPythoN.BbbPred(csv_fp_df,act=True,ret=True,use_bal=True)
# If ret=False the molecules's IDs, predictions, and prediction probabilities are printed.
bbbPythoN.BbbPred(csv_fp_df,ret=False,use_bal=True)bin_bounds:
Contains .csv files specifying the boundaries of bins used to encode descriptor values
into bit strings for every descriptor. bin_bounds_bal.csv, and bin_bounds_imba.csv hold
boundaries based on the balanced, and the imbalanced dataset, respectively.
data:
Contains the different datasets used in "Non-Animal Models for Blood-Brain Barrier Permeability
Evaluation of Drug-Like Compounds".
B3DB_BalTrainTestData.csv: the balanced dataset used to train and validate models.
FeatSel_Data.csv: the dataset used for MI-DSE feature selection.
Inhouse_ImbaTrainTestData.csv: the imbalanced dataset used to train and validate models.
Li_Tong_Wang_ExtVal.csv: the additonal external validation dataset.
desc_names:
Contains mi-dse_descs.csv, and na_descs.csv specifying the names of descriptors chosen by
the MI-DSE feature selection, and the names of descriptors chosen based on their discrepancy
of relative frequency of absence between classes, respectively.
mah_dist_info:
Contains train_act_mah-dist_info_2D+3D_ma_bal, and train_act_mah-dist_info_2D+3D_ma_imba,
as well as train_mah-dist_info_2D+3D_ma_bal, and train_mah-dist_info_2D+3D_ma_imba.
The former two contain information necessary for the compuation of the Mahalanobis distance
between the center of the training set actives and the query molecules.
The latter two contain information necessary for the compuation of the Mahalanobis distance
between the center of the training set and the query molecules.
model:
Contains the two Random Forest classifiers trained on the balanced and imbalanced datasets.
bbbRf_bal.sav: model trained and validated on the balanced dataset.
bbbRf_imba.sav: model trained and validated on the imbalanced dataset.