Building on the Fragments From the Diamond/XChem SARS-CoV-2 Main Protease (MPro) Fragment Screen (Part I)




Like many in the community, I've been spending time looking at the results of a fragment screen of the SARS-CoV-2 main protease (MPro) by the group at the Diamond Light Source.  I thought it might be useful to share some of the techniques and the code that I use when working with the results of fragment screens.  One of the first things I like to do is to virtually expand the fragment by adding a set of substituents and looking for ways that the fragment can be expanded to make new interactions and increase potency.

Once I've added the substituents, I'll generate 3D conformers for the resulting expanded fragments.  The conformers are constrained so that the coordinates of the fragment are fixed.  Since we've fixed the coordinates of the core fragment, we can trivially place these expanded fragments in the original binding site and evaluate their interactions.  This approach is similar to the one we described in a 2004 paper in J. Med. Chem. In this post, I'll describe one approach to computationally expanding a fragment hit.  In a subsequent post, I'll detail the method for generating the 3D conformations and evaluating these conformations in the binding site.

I recently found this preprint and the accompanying GitHub repo from Pavel Polishchuk describing a software tool called CReM (Chemically Reasonable Mutations). The CReM software is like a Swiss army knife for molecular modification and provides several useful capabilities for growing, mutating, and linking molecules. In this post, we'll take a look at a few ways that CReM can be used to modify one of the fragments from the Diamond COVID-19 structures. All of the code and data accompanying this post is available in a Jupyter notebook on GitHub.

CReM takes a simple but effective approach to modifying molecules. The workflow begins by creating a database of matched molecular pairs based on a set of input molecules. The CReM distribution contains a couple of databases compiled from the ChEMBL database. Each matched molecular pair contains a context that specifies an atom environment around which the matched molecular pair was found. When replacing or attaching functional groups, CReM matches the context around the attachment point and uses this context to identify relevant functional groups. The user can specify a radius for the context.  Larger values for this radius specify a more specific context (fewer replacements) while small values specify a less specific environment (more replacements). As mentioned above, CReM provides the ability to mutate, grow, and link molecules. In this post, we'll focus on the "grow" capability, since this provides an excellent way to explore around fragment hits.

We'll start out with this fragment, labeled x0161, in the Diamond set.



To begin, we will create an RDKit molecule from the fragment and generate 2D coordinates.

frag_smi = "COC(=O)C=1C=CC(=CC1)S(=O)(=O)N"
frag_mol = Chem.MolFromSmiles(frag_smi)
AllChem.Compute2DCoords(frag_mol)

Next, we have to specify the location of the database that CReM will use.

crem_db_file = '/Users/pwalters/crem/replacements_sc2.5.db'

I found that some of the substituents that CReM uses contained undesirable (reactive, potentially toxic) functional groups.  In order to remove these, I defined a REOS (Rapid Elimination Of Swill) class that uses some of the code and ideas I laid out in my post on Filtering Chemical Libraries.  The REOS class is initialized with a set of filtering rules.

reos = REOS("bms_filters.csv")

Now that our setup is complete, we can  add substituents to our fragment.

grow_list = list(grow_mol(frag_mol, db_name=crem_db_file,max_atoms=8,return_mol=True))
mol_list = [RemoveHs(x[1]) for x in grow_list]
_ = [AllChem.GenerateDepictionMatching2DStructure(m,frag_mol) for m in mol_list]

Note that I performed two operations on the resulting molecules to improve the aesthetics and make it easier to view the molecules.

  • Remove explicit hydrogens
  • Align the molecules to the fragment so that they're all oriented the same way 
Let's grab a random sample of the molecules we generated and look at them. 


random.shuffle(mol_list)
MolsToGridImage(mol_list[:25],molsPerRow=5)

There are several interesting ideas here, but this also points to the fact that we probably need to add a couple of additional functional group filters. In particular, we need to add a filter for long alkyl chains. We may also be able to get at this by adding a filter for rotatable bonds. 

One of the other cool features in CReM is the ability to exclude certain atoms from expansion or to only expand from a predefined set of atoms.  For instance, let's assume that the sulfonamide NH2 in the fragment is making a hydrogen bond, and we don't want to substitute that part of the molecule. The grow_mol function allows us to specify the atom numbers for the atoms that should not be substituted. In order to use this, we need to specify the atom numbers for the atoms that should not be substituted. To make this easier to use, we'll define a function to return atom numbers matching a SMARTS pattern.  

def get_matching_atoms(mol_in,smarts):
    pat = Chem.MolFromSmarts(smarts)
    return [x[0] for x in mol_in.GetSubstructMatches(pat)]

We can now use this function to get the atom number for the sulfonamide nitrogen.

sulfonamide_N = get_matching_atoms(frag_mol,"[NH2]")



This time we'll add substituents to the molecule, but avoid the sulfonamide nitrogen.


grow_list = list(grow_mol(frag_mol, db_name=crem_db_file,max_atoms=8,protected_ids=sulfonamide_N,return_mol=True))
mol_list = [x for x in mol_list if reos.eval_mol(x)]
mol_list = [RemoveHs(x[1]) for x in grow_list]
_ = [AllChem.GenerateDepictionMatching2DStructure(m,core) for m in mol_list]

MolsToGridImage(mol_list[:25],molsPerRow=5)

Note that in this case, the sulfonamide is not substituted. 

Finally, we can also specify only the parts of the molecule that will be substituted.  Let's say that we only want to substitute the aromatic carbons on the fragment. 

aromatic_atoms = get_matching_atoms(frag_mol,"[cH]")


grow_list = list(grow_mol(frag_mol, db_name=crem_db_file,max_atoms=5,replace_ids=aromatic_atoms,return_mol=True))
mol_list = [x for x in mol_list if reos.eval_mol(x)]
mol_list = [RemoveHs(x[1]) for x in grow_list]
_ = [AllChem.GenerateDepictionMatching2DStructure(m,core) for m in mol_list]

MolsToGridImage(mol_list[:25],molsPerRow=5)

This is just a quick overview of how an Open Source package can be used to generate ideas from hits coming out of a fragment screen.  Hopefully someone will find this useful.  In the next post I'll detail some of the tools that can be used to assess these expanded fragments. 










Comments

Post a Comment

Popular posts from this blog

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

Generative Molecular Design Isn't As Easy As People Make It Look

Image
I was taken aback by a recent CNBC article entitled “ Generative AI will be designing new drugs all on its own in the near future ”.  I should know better than to pay attention to AI articles in the popular press, but I feel that even scientists working in drug discovery may have a skewed perception of what generative AI can and can’t do.  To understand exactly what’s involved, it might be instructive to walk through a typical generative molecular design workflow and point out a few things.  First, these programs are far from autonomous.  Even when presented with a well-defined problem, generative algorithms produce a tremendous amount of nonsense.  Second, domain expertise is essential when sifting through the molecules produced by a generative algorithm.  Without a significant medicinal chemistry background, one can’t make sense of the results.  Third, while a few nuggets exist in the generative modeling output, a lot of work and good old-fashioned cheminformatics are required to ext
Read more

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

Image
Picking up where we left off in Part I , this post covers several other ML in drug discovery topics that interested me in 2023.  Some areas, like large language models, are new, and most of the work is at the proof-of-concept stage.  Others, like active learning, are more mature, and several groups are starting to explore nuances of the methods.   Here’s the structure of Part II.  4. Large Language Models 5. Active Learning 6. Federated Learning 7. Generative Models 8. Explainable AI 9. Other Stuff 4. Large Language Models The emergence of GPT-4 and ChatGPT brought considerable attention to large language models (LLMs) in 2023.  In November and December, several large pharmas held “AI Day” presentations featuring LLM applications for clinical trial data analysis. Many of these groups demonstrated the ability of LLMs to ingest large bodies of unstructured clinical data and subsequently generate tables and reports based on natural language queries.  Aside from some very brief demos on co
Read more