We will use our expertise in AI/ML, cheminformatics, structure-based drug design (SBDD), medicinal chemistry to generate hits for NSP3 Macrodomain (Mac1). Using our in-house drug discovery & cheminformatics platform (published in scientific literature, proprietary code), we will identify a suitable subset of compounds from the Enamine Real Database using various filters which follow medicinal chemistry standards & CACHE white paper guidelines. We will implement two deep learning guided approaches to generate hit ideations, generated hits will be considered as template for sub-structure/similarity search to generate hit list.
Strategy 1 (AI-guided scaffold hopping):
Earlier research efforts (1-4) led to the identification of several sub-micromolar inhibitors which can be potential starting points for scaffold-hopping, to identify novel IP space. We will implement a reinforcement learning approach (5-6) to generate scaffold ideas to be considered as templates for substructure search against the Enamine database. We will generate a potential hit list (bearing the scaffold moieties of our choice) from the substructure search results and those hits will be further filtered for fit using docking studies with Fitted (7-8).
Strategy 2 (Fragment merging/linking):
Prof. Fraser and his team’s efforts led to identification of 160 fragments bearing around 120 diverse scaffolds (2) and they reported 152 Mac1-ligand complex crystal structures from this study. We will encode interaction patterns, binding modes, and scope for fragment-based expansion using these crystal structures and their complexed fragments. We will implement recurrent neural network (RNN)-based generative model (9-10) which is trained to propose plausible linker ideas for a given two input fragments. A major advantage of this generative model is how it can propose linker ideas that satisfy an interaction constraint with additional defined parameters such as. ring size, linker size, and property ranges (chemical space). Generated virtual hits will act as templates for structure search (substructure and similar) against the Enamine databases. Extracted compounds will be further filtered for fit using rigid docking studies with Fitted (7-8). If the generated hits are too dissimilar from the Enamine library, we will consider proposing custom synthesis of these compounds.
Hit selection criteria:
1.) We will carry out the rigid docking studies for both scaffold hopping and fragment-based strategies
2.) We will pick the top-scoring hits ranked based on our docking scoring function.
3.) We will employ a Quantum Mechanics-Based Scoring Function (QMSF) and re-rank the hits based on the calculated relative free energy of binding using a more accurate scoring function than a standard docking function.
4.) We will use the retrospective data (known actives/fragments) and its interaction profile knowledge to shortlist and then re-rank the hits.
5.) Finally, our team will participate in a “hit-picking party”, wherein we will visualize the predicted poses and make a human-based selection following discussion and critique.
6.) We plan to shortlist a total of 100 hits from both strategies (Scaffold Hopping and Fragment merging).
For the hit optimization stage, this workflow will change as follows: we will design the focused library around the potential hit compounds and filter the hits based on the property and structural filters. We will then undertake similar steps outlined above with the new, focused library.
Ultimately, we would like to establish multiple deep learning strategies (scaffold hopping approach and fragment merging strategy) to design potential inhibitors towards Mac1 domain. From each approach, each hit molecule is carefully assessed for its binding mode, key interactions with NSP3 Mac1 domain, and overall fit value. Total of 100 top ranking hits shortlisted from both approaches will be selected for procurement, and testing. "Computational negative controls" may also be selected to support our hypotheses. In line with the SGC/CACHE principles, we will document our research strategy, progress and publicize it for all to follow and reference; we are taking a research-centered focus with this opportunity. We hope the sharing of our findings will help guide future efforts in deep learning guided approaches and in SBDD.