Modeling Solubility

Written by Bharath Ramsundar and Evan Feinberg

Copyright 2016, Stanford University

Computationally predicting molecular solubility through is useful for drug-discovery. In this tutorial, we will use the deepchem library to fit a simple statistical model that predicts the solubility of drug-like compounds. The process of fitting this model involves four steps:

  1. Loading a chemical dataset, consisting of a series of compounds along with aqueous solubility measurements.
  2. Transforming each compound into a feature vector \(v \in \mathbb{R}^n\) comprehensible to statistical learning methods.
  3. Fitting a simple model that maps feature vectors to estimates of aqueous solubility.
  4. Visualizing the results.

We need to load a dataset of estimated aqueous solubility measurements [1] into deepchem. The data is in CSV format and contains SMILES strings, predicted aqueaous solubilities, and a number of extraneous (for our purposes) molecular properties. Here is an example line from the dataset:

C o m p o u n d I D ESOL predicted log solubility in mols per litre M i n i m u m D e g r e e M o l e c u l a r W e i g h t Numbe r of H-Bon d Donor s N u m b e r o f R i n g s Numb er of Rota tabl e Bond s Polar Surface Area m e a s u r e d l o g s o l u b i l i t y i n m o l s p e r l i t r e
b e n z o t h i a z o l e -2.733 2 1 3 5 . 1 9 1 0 2 0 12.89


Most of these fields are not useful for our purposes. The two fields that we will need are the “smiles” field and the “measured log solubility in mols per litre”. The “smiles” field holds a SMILES string [2] that specifies the compound in question. Before we load this data into deepchem, we will load the data into python and do some simple preliminary analysis to gain some intuition for the dataset.

%autoreload 2
%pdb off
Not Currently Working
from import load_from_disk

dataset_file= "../../datasets/delaney-processed.csv"
dataset = load_from_disk(dataset_file)
print("Columns of dataset: %s" % str(dataset.columns.values))
print("Number of examples in dataset: %s" % str(dataset.shape[0]))
ERROR:root:Line magic function %autoreload not found.
Automatic pdb calling has been turned OFF
/home/leswing/anaconda3/envs/deepchem27/lib/python2.7/site-packages/sklearn/ DeprecationWarning: This module was deprecated in version 0.18 in favor of the model_selection module into which all the refactored classes and functions are moved. Also note that the interface of the new CV iterators are different from that of this module. This module will be removed in 0.20.
  "This module will be removed in 0.20.", DeprecationWarning)
Columns of dataset: ['Compound ID' 'ESOL predicted log solubility in mols per litre'
 'Minimum Degree' 'Molecular Weight' 'Number of H-Bond Donors'
 'Number of Rings' 'Number of Rotatable Bonds' 'Polar Surface Area'
 'measured log solubility in mols per litre' 'smiles']
Number of examples in dataset: 1128

To gain a visual understanding of compounds in our dataset, let’s draw them using rdkit. We define a couple of helper functions to get started.

import tempfile
from rdkit import Chem
from rdkit.Chem import Draw
from itertools import islice
from IPython.display import Image, HTML, display

def display_images(filenames):
    """Helper to pretty-print images."""
        ["<img style='width: 140px; margin: 0px; float: left; border: 1px solid black;' src='%s' />"
         % str(s) for s in sorted(filenames)])

def mols_to_pngs(mols, basename="test"):
    """Helper to write RDKit mols to png files."""
    filenames = []
    for i, mol in enumerate(mols):
        filename = "%s%d.png" % (basename, i)
        Draw.MolToFile(mol, filename)
    return filenames

Now, we display some compounds from the dataset:

num_to_display = 14
molecules = []
for _, data in islice(dataset.iterrows(), num_to_display):

Analyzing the distribution of solubilities shows us a nice spread of data.

%matplotlib inline
import matplotlib
import numpy as np
import matplotlib.pyplot as plt

solubilities = np.array(dataset["measured log solubility in mols per litre"])
n, bins, patches = plt.hist(solubilities, 50, facecolor='green', alpha=0.75)
plt.xlabel('Measured log-solubility in mols/liter')
plt.ylabel('Number of compounds')
plt.title(r'Histogram of solubilities')

With our preliminary analysis completed, we return to the original goal of constructing a predictive statistical model of molecular solubility using deepchem. The first step in creating such a molecule is translating each compound into a vectorial format that can be understood by statistical learning techniques. This process is commonly called featurization. deepchem packages a number of commonly used featurization for user convenience. In this tutorial, we will use ECPF4 fingeprints [3].

deepchem offers an object-oriented API for featurization. To get started with featurization, we first construct a Featurizer object. deepchem provides the CircularFingeprint class (a subclass of Featurizer that performs ECFP4 featurization).

import deepchem as dc

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

Now, let’s perform the actual featurization. deepchem provides the CSVLoader class for this purpose. The featurize() method for this class loads data from disk and uses provided Featurizerinstances to transform the provided data into feature vectors.

To perform machine learning upon these datasets, we need to convert the samples into datasets suitable for machine-learning (that is, into data matrix \(X \in \mathbb{R}^{n\times d}\) where \(n\) is the number of samples and \(d\) the dimensionality of the feature vector, and into label vector \(y \in \mathbb{R}^n\)). deepchem provides the Dataset class to facilitate this transformation. This style lends itself easily to validation-set hyperparameter searches, which we illustate below.

loader =
      tasks=["measured log solubility in mols per litre"], smiles_field="smiles",
dataset = loader.featurize(dataset_file)
Loading raw samples now.
shard_size: 8192
About to start loading CSV from ../../datasets/delaney-processed.csv
Loading shard 1 of size 8192.
Featurizing sample 0
Featurizing sample 1000
TIMING: featurizing shard 0 took 2.016 s
TIMING: dataset construction took 2.061 s
Loading dataset from disk.
/home/leswing/Documents/deepchem/deepchem/data/ UnicodeWarning: Unicode equal comparison failed to convert both arguments to Unicode - interpreting them as being unequal
  if y[ind, task] == "":

When constructing statistical models, it’s necessary to separate the provided data into train/test subsets. The train subset is used to learn the statistical model, while the test subset is used to evaluate the learned model. In practice, it’s often useful to elaborate this split further and perform a train/validation/test split. The validation set is used to perform model selection. Proposed models are evaluated on the validation-set, and the best performed model is at the end tested on the test-set.

Choosing the proper method of performing a train/validation/test split can be challenging. Standard machine learning practice is to perform a random split of the data into train/validation/test, but random splits are not well suited for the purposes of chemical informatics. For our predictive models to be useful, we require them to have predictive power in portions of chemical space beyond the set of molecules in the training data. Consequently, our models should use splits of the data that separate compounds in the training set from those in the validation and test-sets. We use Bemis-Murcko scaffolds [5] to perform this separation (all compounds that share an underlying molecular scaffold will be placed into the same split in the train/test/validation split).

splitter = dc.splits.ScaffoldSplitter(dataset_file)
train_dataset, valid_dataset, test_dataset = splitter.train_valid_test_split(
Computing train/valid/test indices
About to generate scaffolds
Generating scaffold 0/1128
Generating scaffold 1000/1128
About to sort in scaffold sets
TIMING: dataset construction took 0.058 s
Loading dataset from disk.
TIMING: dataset construction took 0.027 s
Loading dataset from disk.
TIMING: dataset construction took 0.030 s
Loading dataset from disk.

Let’s visually inspect some of the molecules in the separate splits to verify that they appear structurally dissimilar. The FeaturizedSamples class provides an itersamples method that lets us obtain the underlying compounds in each split.

train_mols = [Chem.MolFromSmiles(compound)
              for compound in train_dataset.ids]
display_images(mols_to_pngs(train_mols, basename="train"))
valid_mols = [Chem.MolFromSmiles(compound)
              for compound in valid_dataset.ids]
display_images(mols_to_pngs(valid_mols, basename="valid"))

Notice the visual distinction between the train/validation splits. The most-common scaffolds are reserved for the train split, with the rarer scaffolds allotted to validation/test.

The performance of common machine-learning algorithms can be very sensitive to preprocessing of the data. One common transformation applied to data is to normalize it to have zero-mean and unit-standard-deviation. We will apply this transformation to the log-solubility (as seen above, the log-solubility ranges from -12 to 2).

transformers = [
    dc.trans.NormalizationTransformer(transform_y=True, dataset=train_dataset)]

for dataset in [train_dataset, valid_dataset, test_dataset]:
  for transformer in transformers:
      dataset = transformer.transform(dataset)
TIMING: dataset construction took 0.042 s
Loading dataset from disk.
TIMING: dataset construction took 0.009 s
Loading dataset from disk.
TIMING: dataset construction took 0.009 s
Loading dataset from disk.

The next step after processing the data is to start fitting simple learning models to our data. deepchem provides a number of machine-learning model classes.

In particular, deepchem provides a convenience class, SklearnModel that wraps any machine-learning model available in scikit-learn [6]. Consequently, we will start by building a simple random-forest regressor that attempts to predict the log-solubility from our computed ECFP4 features. To train the model, we instantiate the SklearnModel object, then call the fit() method on the train_dataset we constructed above. We then save the model to disk.

from sklearn.ensemble import RandomForestRegressor

sklearn_model = RandomForestRegressor(n_estimators=100)
model = dc.models.SklearnModel(sklearn_model)

We next evaluate the model on the validation set to see its predictive power. deepchem provides the Evaluator class to facilitate this process. To evaluate the constructed model object, create a new Evaluator instance and call the compute_model_performance() method.

from deepchem.utils.evaluate import Evaluator

metric = dc.metrics.Metric(dc.metrics.r2_score)
evaluator = Evaluator(model, valid_dataset, transformers)
r2score = evaluator.compute_model_performance([metric])
computed_metrics: [0.15666969194533098]
{'r2_score': 0.15666969194533098}

The performance of this basic random-forest model isn’t very strong. To construct stronger models, let’s attempt to optimize the hyperparameters (choices made in the model-specification) to achieve better performance. For random forests, we can tweak n_estimators which controls the number of trees in the forest, and max_features which controls the number of features to consider when performing a split. We now build a series of SklearnModels with different choices for n_estimators and max_features and evaluate performance on the validation set.

def rf_model_builder(model_params, model_dir):
  sklearn_model = RandomForestRegressor(**model_params)
  return dc.models.SklearnModel(sklearn_model, model_dir)
params_dict = {
    "n_estimators": [10, 100],
    "max_features": ["auto", "sqrt", "log2", None],

metric = dc.metrics.Metric(dc.metrics.r2_score)
optimizer = dc.hyper.HyperparamOpt(rf_model_builder)
best_rf, best_rf_hyperparams, all_rf_results = optimizer.hyperparam_search(
    params_dict, train_dataset, valid_dataset, transformers,
Fitting model 1/8
hyperparameters: {'n_estimators': 10, 'max_features': 'auto'}
computed_metrics: [0.16084049792955879]
Model 1/8, Metric r2_score, Validation set 0: 0.160840
    best_validation_score so far: 0.160840
Fitting model 2/8
hyperparameters: {'n_estimators': 10, 'max_features': 'sqrt'}
computed_metrics: [0.24188860561412873]
Model 2/8, Metric r2_score, Validation set 1: 0.241889
    best_validation_score so far: 0.241889
Fitting model 3/8
hyperparameters: {'n_estimators': 10, 'max_features': 'log2'}
computed_metrics: [0.13923134159755857]
Model 3/8, Metric r2_score, Validation set 2: 0.139231
    best_validation_score so far: 0.241889
Fitting model 4/8
hyperparameters: {'n_estimators': 10, 'max_features': None}
computed_metrics: [0.068066979889086832]
Model 4/8, Metric r2_score, Validation set 3: 0.068067
    best_validation_score so far: 0.241889
Fitting model 5/8
hyperparameters: {'n_estimators': 100, 'max_features': 'auto'}
computed_metrics: [0.13350083343444363]
Model 5/8, Metric r2_score, Validation set 4: 0.133501
    best_validation_score so far: 0.241889
Fitting model 6/8
hyperparameters: {'n_estimators': 100, 'max_features': 'sqrt'}
computed_metrics: [0.27500018179983798]
Model 6/8, Metric r2_score, Validation set 5: 0.275000
    best_validation_score so far: 0.275000
Fitting model 7/8
hyperparameters: {'n_estimators': 100, 'max_features': 'log2'}
computed_metrics: [0.24719988102520696]
Model 7/8, Metric r2_score, Validation set 6: 0.247200
    best_validation_score so far: 0.275000
Fitting model 8/8
hyperparameters: {'n_estimators': 100, 'max_features': None}
computed_metrics: [0.15709535184995427]
Model 8/8, Metric r2_score, Validation set 7: 0.157095
    best_validation_score so far: 0.275000
computed_metrics: [0.94421662988139621]
Best hyperparameters: (100, 'sqrt')
train_score: 0.944217
validation_score: 0.275000

The best model achieves significantly higher \(R^2\) on the validation set than the first model we constructed. Now, let’s perform the same sort of hyperparameter search, but with a simple deep-network instead.

import numpy.random

params_dict = {"learning_rate": np.power(10., np.random.uniform(-5, -3, size=1)),
               "decay": np.power(10, np.random.uniform(-6, -4, size=1)),
               "nb_epoch": [20] }
n_features = train_dataset.get_data_shape()[0]
def model_builder(model_params, model_dir):
  model = dc.models.TensorflowMultiTaskRegressor(
    1, n_features, layer_sizes=[1000], dropouts=[.25],
    batch_size=50, **model_params)
  return model

optimizer = dc.hyper.HyperparamOpt(model_builder)
best_dnn, best_dnn_hyperparams, all_dnn_results = optimizer.hyperparam_search(
    params_dict, train_dataset, valid_dataset, transformers,
Fitting model 1/1
hyperparameters: {'learning_rate': 0.00023048114774615143, 'nb_epoch': 20, 'decay': 4.6386701591998651e-06}
Training for 20 epochs
On batch 0
Ending epoch 0: Average loss 2.90304
On batch 0
Ending epoch 1: Average loss 1.80178
On batch 0
Ending epoch 2: Average loss 1.49158
On batch 0
Ending epoch 3: Average loss 1.32873
On batch 0
Ending epoch 4: Average loss 1.1639
On batch 0
Ending epoch 5: Average loss 1.00606
On batch 0
Ending epoch 6: Average loss 0.93537
On batch 0
Ending epoch 7: Average loss 0.866738
On batch 0
Ending epoch 8: Average loss 0.785585
On batch 0
Ending epoch 9: Average loss 0.735722
On batch 0
Ending epoch 10: Average loss 0.693006
On batch 0
Ending epoch 11: Average loss 0.620769
On batch 0
Ending epoch 12: Average loss 0.59393
On batch 0
Ending epoch 13: Average loss 0.565747
On batch 0
Ending epoch 14: Average loss 0.558274
On batch 0
Ending epoch 15: Average loss 0.522325
On batch 0
Ending epoch 16: Average loss 0.52158
On batch 0
Ending epoch 17: Average loss 0.483039
On batch 0
Ending epoch 18: Average loss 0.472697
On batch 0
Ending epoch 19: Average loss 0.463411
On batch 0
TIMING: model fitting took 4.142 s True
computed_metrics: [0.19197438493808416]
Model 1/1, Metric r2_score, Validation set 0: 0.191974
    best_validation_score so far: 0.191974
computed_metrics: [0.81127373811679659]
Best hyperparameters: (0.00023048114774615143, 20, 4.6386701591998651e-06)
train_score: 0.811274
validation_score: 0.191974

Now that we have a reasonable choice of hyperparameters, let’s evaluate the performance of our best models on the test-set.

rf_test_evaluator = Evaluator(best_rf, test_dataset, transformers)
rf_test_r2score = rf_test_evaluator.compute_model_performance([metric])
print("RF Test set R^2 %f" % (rf_test_r2score["r2_score"]))
computed_metrics: [0.36485280404796627]
RF Test set R^2 0.364853
dnn_test_evaluator = Evaluator(best_dnn, test_dataset, transformers)
dnn_test_r2score = dnn_test_evaluator.compute_model_performance([metric])
print("DNN Test set R^2 %f" % (dnn_test_r2score["r2_score"]))
computed_metrics: [0.24496914941196501]
DNN Test set R^2 0.244969

Now, let’s plot the predicted \(R^2\) scores versus the true \(R^2\) scores for the constructed model.

task = "measured log solubility in mols per litre"
predicted_test = best_rf.predict(test_dataset)
true_test = test_dataset.y
plt.scatter(predicted_test, true_test)
plt.xlabel('Predicted log-solubility in mols/liter')
plt.ylabel('True log-solubility in mols/liter')
plt.title(r'RF- predicted vs. true log-solubilities')
task = "measured log solubility in mols per litre"
predicted_test = best_dnn.predict(test_dataset)
true_test = test_dataset.y
plt.scatter(predicted_test, true_test)
plt.xlabel('Predicted log-solubility in mols/liter')
plt.ylabel('True log-solubility in mols/liter')
plt.title(r'DNN predicted vs. true log-solubilities')

[1] John S. Delaney. ESOL: Estimating aqueous solubility directly from molecular structure. Journal of Chemical Information and Computer Sciences, 44(3):1000–1005, 2004.

[2] Anderson, Eric, Gilman D. Veith, and David Weininger. SMILES, a line notation and computerized interpreter for chemical structures. US Environmental Protection Agency, Environmental Research Laboratory, 1987.

[3] Rogers, David, and Mathew Hahn. “Extended-connectivity fingerprints.” Journal of chemical information and modeling 50.5 (2010): 742-754.

[4] Van Der Walt, Stefan, S. Chris Colbert, and Gael Varoquaux. “The NumPy array:a structure for efficient numerical computation.” Computing in Science & Engineering 13.2 (2011): 22-30.

[5] Bemis, Guy W., and Mark A. Murcko. “The properties of known drugs. 1. Molecular frameworks.” Journal of medicinal chemistry 39.15 (1996): 2887-2893.

[6] Pedregosa, Fabian, et al. “Scikit-learn: Machine learning in Python.” The Journal of Machine Learning Research 12 (2011): 2825-2830.