Using Reaction Transforms to Understand SAR

One of the most effective ways of understanding structure-activity relationships (SAR) is by comparing pairs of compounds which differ by a single, consistent feature.  By doing this, we can often understand the impact of this feature on biological activity or physical properties.  In order to effectively identify these pairs, we sometimes need an automated method that can sift through large datasets.  In this post, we will show how we can use the chemical reaction capabilities in a Cheminformatics toolkit to identify interesting pairs of compounds.   As usual, the code to accompany this post is on GitHub.

The idea of comparing the biological activity of pairs of compounds which only differ by a single feature has been a key concept since the beginning of medicinal chemistry.  Over the last 15 years, software tools for generating "matched molecular pairs" (MMP) have become a common component of Cheminformatic analyses.  For those unfamiliar with the technique, MMP analysis uses algorithms to identify pairs of molecules which differ by a single feature.  For example, in the pairs below, the imidazopyridazine (IP) scaffold on the left has been replaced by a triazolopyridine (TP) scaffold on the right.  Note that while the scaffold has changed, the substituents are the same.


Let's say that we have a large dataset and we want to identify all of the pairs where the scaffold has changed from IP to TP, but everything else is the same.

We could run a standard matched molecular pair analysis, but there are two downsides to this.

  • Many matched pair analysis methods are good at identifying changes in substituents, but sometimes fail to detect scaffold changes. 
  • Matched pair analysis techniques often identify dozens or even hundreds of pairs.  We then have to sift through these pairs to find what we're looking for.  
In this case, we know the pair that we're looking for, so why not look for exactly what we want.  One way to do this is to define a reaction transform that will convert IP to TP.   We can do this by first creating an atom mapped reaction transform.

This reaction transform, which can be created in most chemical drawing programs (ChemDraw, Marvin Sketch, etc) defines the mapping between atoms in the reactant (left) and the product(right).   With this reaction transform in hand, we can simply apply the transform to every molecule in our dataset.  If the molecule contains the IP scaffold on the left, it will be transformed into the TP scaffold on the right.  We can then check to see if the "product" molecule is in the original dataset.  If it is, we have a pair.  We can then collect these pairs and examine their differences in activity.  

As an example let's consider this 2012 paper from Bioorganic & Medicinal Chemistry Letters, that discusses the hit to lead evaluation of a set of PIM kinase inhibitors.  The ChEMBL database contains 57 structures from this paper, along with the corresponding PIM1 IC50 values.  It would be useful to be able to determine the impact of the scaffold change on the PIM1 IC50.   I've created a Git repository with the code for identifying the pairs, as well as a Jupyter notebook that demonstrates how to visualize the output.   

To perform the analysis on the dataset in the repo, we can execute this command:

transformer_search.py --rxn rxn.rxn --in CHEMBL1949661.csv --out out.sdf

The output contains alternating pairs of molecules where the first contains the IP scaffold and the next contains the TP scaffold.  The Jupyter notebook shows how we can then create a plot that compares the IC50 values for the two scaffolds.   We can see that, for the most part, the IP scaffold is more active, but there are exceptions. 

Of course, there's lot's more we could do, including making this plot interactive.  We'll save that for another day.  I'd like to thank Emanuele Perola for help in debugging the code and Greg Landrum and Lukas Pravda for sharing some RDKit tricks. 






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