Viewing Clustered Chemical Structures in a Jupyter Notebook

In Cheminformatics, we frequently run into cases where we want to look at leader/follower relationships between chemical structures.  For instance, if we've clustered a set of molecules, we might want to start by looking at a table with one example structure for each cluster.  We'd then like to be able to select one or more "interesting" clusters and drill down to the cluster members.  While this is a frequent workflow, I'm not aware of commercially or freely available tools that do a great job of supporting the exploration of leader/follower relationships with chemical structures.  In this post, we'll look at one way of connecting a couple of open source libraries to view cluster representatives and cluster members.  As usual, the code and data associated with this post are available on GitHub. 

Rather than trying to explain this more fully, let's consider an example.  In this example, we'll look at a set of 1,495 drugs from the ChEMBL database.  If we cluster the structures using the Butina clustering algorithm, as implemented in the RDKit, we'll end up with each structure in our dataset having a cluster identifier.  


We can use the Pandas groupby function to group the dataframe by cluster and select the first cluster member as an example.  If we wanted to, we could get a bit fancier and select the molecule closest to the cluster center as the representative example. 

Now that we have our cluster representatives in one dataframe, we can display that dataframe using the nifty mols2grid component. In the grid below we're showing a single representative for each cluster, with the number of molecules in the cluster shown below the structure. In addition to displaying a grid of structures, mols2grid has checkboxes that allow the user to select structures.   These selections can then be accessed using mols2grid.selection.  



Since we can query the mols2grid object to get the selected structure(s), we can use this information to create a new dataframe containing only the selected cluster(s).  I've seen a lot of people on Twitter complain about indexing in Pandas.  I agree that indexing can be confusing. That's why I tend to use the Pandas query facility, rather than indexing, whenever possible. 

sel = mols2grid.selection.keys()
sel_df = df.query("Cluster in @sel")

Now that we have a dataframe with the selected cluster(s) we can display the molecules from the selected cluster(s) in another table.  I wanted to be able to use another mols2grid instance to show the molecules from the selected clusters(s).  Unfortunately, that won't work.  The mols2grid component maintains global state.  If we instantiate a new mols2grid object, we lose the state of the original mols2grid object. It seems like there may be a way around this by modifying the mols2grid code and storing the state in a list or a dictionary.  That's a project for another day. 

We can't create a second mols2grid instance but we can use the MolsToGridImage function from the RDKit to show the molecules from the selected cluster. 

mol_list = [Chem.MolFromSmiles(x) for x in sel_df.SMILES]
legends = list(sel_df.Name.astype(str))
Draw.MolsToGridImage(mol_list,molsPerRow=5,legends=legends,useSVG=True)

One thing I'm really picky about when viewing molecules is having similar structures aligned.  I don't like the mental gymnastics required to imagine how structures align.  Fortunately, we can use the maximum common subgraph (MCS) implementation in the RDKit to align the structures. 

mcs=rdFMCS.FindMCS(mol_list)
mcs_query = Chem.MolFromSmarts(mcs.smartsString)            
AllChem.Compute2DCoords(mcs_query)
for m in mol_list: 
    AllChem.GenerateDepictionMatching2DStructure(m,mcs_query)

Now that we have the pieces in place, we have a two-step workflow for viewing cluster representatives and cluster members.  
1. Click on the checkbox in the mols2grid component to select the cluster(s) to view. 
2. Shift-return on the cell below the mols2grid component to view the cluster members. 


This workflow is ok, but it's not a smooth as I'd like it to be.  I'm eager to hear any ideas on how to improve it.  I'd also be interested in hearing approaches that others have for examining leader/follower datasets like this. 


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