How to get ScaffoldSplitter output as a CSV file?

#1

Hello,

I am a newbie to python/deepchem. I need to do a scaffold split on my own dataset (to evaluate ROCS scaffold hopping).

I tried running the example and I am able to get a split. But I don’t know as to how to write the output as a csv file. Any help is appreciated.

This is what I did so far with example dataset (example.csv) provided in deepchem.

import deepchem as dc

import pandas as pd

import os

current_dir=os.path.dirname(os.path.realpath(‘file’))

input_data=os.path.join(current_dir,’./example.csv’)

tasks=[‘log-solubility’]

featurizer=dc.feat.CircularFingerprint(size=1024)

loader = dc.data.CSVLoader(tasks=tasks, smiles_field=“smiles”,featurizer=featurizer)

dataset=loader.featurize(input_data)

splitter = dc.splits.ScaffoldSplitter(input_file)

train_dataset, test_dataset = splitter.train_test_split(dataset)

len(train_dataset),len(test_dataset)

The output I am trying to get are two csv files train_dataset and test_dataset. Ideally with Compound ID.

Thanks
Mohamed

#2

This is unfortunately not as straightforward as it should be. There’s an open GitHub issue discussing how to make this easier. Here’s how you can do this:

train_df = pd.DataFrame(train_dataset.ids, columns=["smiles"])
test_df = pd.DataFrame(test_dataset.ids, columns=["smiles"])
train_df.to_csv("/tmp/train.csv")
test_df.to_csv("/tmp/test.csv")

Your code example didn’t load a compound ID. If you did have a compound ID for compounds, and you used unique isomeric smiles, you could maintain a dictionary mapping smiles to IDs and recover your IDs from the smiles.

1 Like
#3

Thanks a lot, Bharath! It’s very helpful.

1 Like