Compound Ideation: Fine-Grained Binding Pocket Exploration Through Ligand Mutation and Fragment Growing

Compound Ideation:
Fine-Grained Binding Pocket Exploration Through Ligand Mutation and Fragment Growing

In the initial stages of early drug discovery, the focus is often on compound ideation: Investigating the Chemical Space around a compound of interest via structure-guided screening for decorations and modifications.
This includes the assessment of possible modifications to improve the physicochemical properties of the drug candidate and/or the addition of decorations forming high-quality interactions with the target to improve the potency of the compound.

In this application spotlight, we used the tools MedChemesis and FastGrow, both part of our drug design dashboard SeeSAR, on several drug discovery scenarios. We highlight their potential as valuable resources for ideation of improved drug candidates in the context of predicted potency, ligand-lipophilicity efficiency (LLE), and important ADME properties such as logP and logS.

MedChemesis — Transformations Driven by Medicinal Chemistry

The all-new MedChemesis employs 290 commonly used medicinal chemistry transformations and applies these to a ligand-target complex in seconds. Automatically, you identify favorable modifications with neither extended mouse clicking nor exhaustive library enumeration of all possibilities. In Case 1 and 2, we examined the properties of proposed results after a single MedChemesis run (max. 100 outputs, see Figure 1).

In Case 1, MedChemesis suggested a total of 81 compounds, with 45 of them (56%) showing enhanced predicted binding affinity. Meanwhile, in Case 2, a set of 65 compounds was generated, and 31 of these molecules exhibited improved predicted affinity (48%). Several candidates displayed additional H-bond interactions within the target's binding site, providing valuable suggestions for the compound ideation process.

Furthermore, MedChemesis identified three and two potential growing opportunities for Case 1 (A-C) and 2 (D and E), respectively. The growing sites were subsequently filled with FastGrow in the section below.

Fine-Tuning ADME Properties

Figure 1. Distribution of MedChemesis results for Case 1 and 2.

As for the ADME properties: MedChemesis results displayed a broad distribution in for the LLE and logP parameters, with several candidates in the desired range. For logS, remarkably, results for Case 1 exhibited increased logS values mainly driven by the nature of MedChemesis to find modifications that also increase the predicted binding affinity, leading to several compounds containing halogen decorations to complement the lipophilic binding site.

FastGrow — Augmenting Fragment-Based Drug Discovery (FBDD) and Scaffold Hopping

Growing from a fragment to satisfy unoccupied binding cavities has emerged as a state-of-the-art method to improve a binder’s potency.
While full enumeration of fragments with a diverse set of molecular decorations followed by template-based docking is common, it is slow and therefore lacks the possibility to quickly probe new hypotheses and do on-the-fly exploration.

The new FastGrow algorithm rapidly explores conformations of over 120K fragments in seconds on standard hardware for best shape-complementing extensions, while optimizing the H-bonding. Distribution of results after respective growing runs (max. 500 outputs each) for Case 1 (A-C) and 2 (D and E) are shown in Figure 2

Exploration of the Nearest Chemical Space Around a Compound

Figure 2. Distribution of FastGrow results for Case 1 and 2.

To mimic the modus operandi of a computational or medicinal chemists in a research setting, a maximum of 500 FastGrow results were generated. The generated molecule set was then subsequently filtered for compounds with a molecular weight <500 to focus on classical drug-like molecules. Ultimately, this resulted in 52, 336 and 288 extended molecules for Case 1 (growing sites A to C, respectively), and 329 and 70 molecules for Case 2 (growing sites D and E, respectively). Out of those results, 25 (48%), 286 (85%), 248 (86%), 50 (15%) and 27 (39%) displayed improved predicted binding affinity.
Since FastGrow screens for the best compound decorations based on binding site shape complementarity and placement of interaction partners, the whole binding site is sampled during the process, typically unveiling a plethora of solutions for follow-up.

Again, a broad distribution of ADME properties highlights the chemical diversity of the generated molecules.

Side Chain Replacement in a Lead Structure

Figure 3. Distribution of FastGrow results for Case 3 and 4.

The nature of FastGrow can also be exploited to replace (potentially unwanted) side chains of a ligand. Furthermore, it provides novel ideas for interactions within the binding site and explores structures with improved properties.
In Case 3 and Case 4, a considerable portion of the ligand was selected to be swapped by suggested side chains, following the filtering mentioned above (results presented in Figure 3.

For Case 3, 290 results were obtained, of which 214 (74%) displayed improved predicted binding affinity. For Case 4, 411 results were generated with 236 (57%) potentially improved candidates.

Summary

Both tools, MedChemesis and FastGrow, uncover unexpected distributions and generate numerous novel compound proposals with improved predicted binding. Tailored to specific cases, they enable ultrafast exploration with an emphasis on ADME, yielding promising candidates.

Additonally, pharmacophore constraints can be applied within the software to fine tune results, while synergy with other methods enhances ideation and design. The approaches also support the creation of molecule subsets for other methods like machine learning and all sorts of generative AI.

Try MedChemesis and FastGrow for your projects, discover novel solutions you wouldn't have considered otherwise!

SeeSAR