Visualizing Decision Trees

A 2016 paper by Wicker and Cooper, describing a molecular descriptor designed to capture molecular flexibility, popped up on Twitter this week.  This paper reminded me of the power of a simple decision tree.  Decision trees can often provide an efficient way of looking at the relationship between molecular descriptors and experimental data.   They can also provide a means of understanding the relationship between sets of experiments, particularly with pharmacokinetic data.

In this spirit, I thought I'd put together a quick post showing how to build and visualize a decision tree.  This post will also show off a couple of useful Python libraries that I've recently integrated into my workflow. 

In their paper, Wicker and Cooper use a set of 40,541 commercially available molecules, from the ZINC database, to establish a relationship between molecular flexibility and the ability of a molecule to crystallize.  The dataset is divided into two subsets.
  • “observed to crystallize” - molecules that occur in both ZINC and the Cambridge Crystallographic Database (CSD).  
  • “not observed to crystallize” - molecules found in ZINC but not in the CSD
The assumption here is that the molecules not found in the CSD were difficult to crystallize.  There could, of course, be a number of other reasons the “not observed to crystallize” set was not in the CSD, but let's just go with it. 

Time to write some code.  Let's start by importing the necessary Python libraries. 
import pandas as pd
import janitor
from sklearn import tree
from dtreeviz.trees import *
from sklearn.metrics import matthews_corrcoef

Here's a quick rundown on the libraries.
  • pandas - the go-to library for anything having to do with tabular data
  • janitor - the amazing PyJanitor library for cleaning data, more on this below
  • sklearn - the scikit-learn library for all things machine learning 
  • dtreeviz - cool visualizations of decision trees, more below
We start by reading the training set. 
train_df = pd.read_csv("train_desc_with_names.csv")
Let's take a look at the names of the columns in the dataframe.
train_df.columns
Index(['Unnamed: 0', 'MolWt', 'HeavyAtomMolWt', 'NumRadicalElectrons',
       'NumValenceElectrons', 'BalabanJ', 'BertzCT', 'Chi0', 'Chi0n', 'Chi0v',
       ...
       'R6', 'R7', 'R8', 'R9', 'RG10', 'AFRC', 'BFRC', 'nConf20', 'Label',
       'Label names'],
      dtype='object', length=223)
Note that some of the column names have spaces, which can be somewhat inconvenient. Fortunately, the PyJanitor library has functions that enable us to do all sorts of data cleaning. We can use the "clean_names" function to remove the spaces from the column names. We are also taking advantage of the very cool method chaining capability provided by PyJanitor. 
train_df = (train_df.clean_names(case_type="preserve")
            .remove_columns(["Unnamed_0","Label_names"]))
Next, we can extract the x variables (descriptors) and the y variables (response) from the dataframe.
train_X = train_df.remove_columns(["Label"])
train_y = train_df.Label
Time to build the model, let's instantiate a DecisionTreeClassifier.
cls = tree.DecisionTreeClassifier(max_depth=2)
Next, we can train the classifier.
cls.fit(train_X,train_y)
DecisionTreeClassifier(ccp_alpha=0.0, class_weight=None, criterion='gini',
                       max_depth=2, max_features=None, max_leaf_nodes=None,
                       min_impurity_decrease=0.0, min_impurity_split=None,
                       min_samples_leaf=1, min_samples_split=2,
                       min_weight_fraction_leaf=0.0, presort='deprecated',
                       random_state=None, splitter='best')

Now that we have the model, we can use the dtreeviz package to create a visualization.  The scikit-learn package has a method for displaying decision trees (see Figure 6 in Wicker and Cooper for an example), but I haven't found it to be particularly useful.  I like the visualization in dtreeviz because it shows the relationships between the distributions and the decision tree.


feature_names = list(train_X.columns)
viz = dtreeviz(cls, train_X, train_y, feature_names = feature_names, 
               target_name = "Crystallize",class_names=["No","Yes"],scale=2)
viz









The plot above shows the decision points on the x-axis of the histogram showing the distribution.  The distributions in the terminal nodes are shown as pie charts.



It looks like the model is performing well, let's take a look at the performance on the test set.  First, we'll read the data and clean it with PyJanitor.


test_df = pd.read_csv("test_desc_with_names.csv")
test_df = (test_df.clean_names(case_type="preserve")
        .remove_columns(["Unnamed_0","Label_names"]))
test_X = test_df.remove_columns(["Label"])
test_y = test_df.Label

Now we can check the Matthews Correlation Coefficient for the test set.


pred = cls.predict(test_X)
matthews_corrcoef(test_y, pred)


0.7781038157491144

Not bad for a very simple model.  While it's true that there are a lot of sophisticated machine learning models out there, the humble decision tree can often provide an easy way of creating interpretable models.   The dtreeviz package is a nice add-on that enables you to quickly see what the decision tree is doing.  Hopefully, you'll find this useful and integrate it into your data analysis workflow.  






As usual, the code for this post is available in a Jupyter notebook on GitHub.  If you want to run the code without installing software, there is a version on Binder.












































Comments











Post a Comment























Popular posts from this blog









We Need Better Benchmarks for Machine Learning in Drug Discovery




















Image







Most papers describing new methods for machine learning (ML) in drug discovery report some sort of benchmark comparing their algorithm and/or molecular representation with the current state of the art.  In the past, I’ve written extensively about statistics  and how methods should be compared .  In this post, I’d like to focus instead on the datasets we use to benchmark and compare methods.  Many papers I’ve read recently use the MoleculeNet dataset, released by the Pande group at Stanford in 2017, as the “standard” benchmark.   This is a mistake.  In this post, I’d like to use the MoleculeNet dataset to point out flaws in several widely used benchmarks.  Beyond this, I’d like to propose some alternate strategies that could be used to improve benchmarking efforts and help the field to move forward.   To begin, let’s examine the MoleculeNet benchmark, which to date, has been cited more than 1,800 times.   The MoleculeNet collection consists of 16 datasets divided into 4 categories.  Qua








Read more










 AI in Drug Discovery 2023 - A Highly Opinionated Literature Review (Part I)




















Image







Here’s the first part of my review of some interesting machine learning (ML) papers I read in 2023.  As with the previous  editions , this shouldn’t be considered a comprehensive review.  The papers covered here reflect my research interests and biases, and I’ve certainly overlooked areas that others consider vital.  This post is pretty long, so I've split it into three parts, with parts II and III to be posted in the next couple of weeks.    I. Docking, protein structure prediction, and benchmarking II. Large Language Models, active learning, federated learning, generative models, and explainable AI III. Review articles 2023 was a bit of a mixed bag for AI in drug discovery.  Several groups reported that the deep learning methods for protein-ligand docking weren’t quite what they were initially cracked up to be.  AlphaFold2 became pervasive, and people started to investigate, with mixed success, the utility of predicted protein structures.  There were reports of significant advanc








Read more










Getting Real with Molecular Property Prediction




















Image







Introduction If you believe everything you read in the popular press, this AI business is easy. Just ask ChatGPT, and the perfect solution magically appears. Unfortunately, that's not the reality. In this post, I'll walk through a predictive modeling example and demonstrate that there are still a lot of subtleties to consider. In addition, I want to show that data is critical to building good machine learning (ML) models. If you don't have the appropriate data, a simple empirical approach may be better than an ML model.  A recent paper from Cheng Fang and coworkers at Biogen  presents prospective evaluations of machine learning models on several ADME endpoints.  As part of this evaluation, the Biogen team released a large dataset of measured in vitro  assay values for several thousand commercially available compounds.  One component of this dataset is 2,173 solubility values measured at pH 6.8 using chemiluminescent nitrogen detection  (CLND), a technique currently consider








Read more