Our approach called DockAI is a new technology that combines docking with a state-of-the-art active learning methodology to significantly improve the efficiency and effectiveness of virtual screening and hit identification.

Our approach combines the expertise of Kozakov Lab at Stony Brook and Tropsha Lab at UNC. Our workflow uses several complimentary modules for identification of high affinity hits for a given protein target with a known 3D structure. Identification of binding site hot-spot 4(8) information together with conventional structure-based virtual screening methods are enabling components of our hit selection approach.

Glide by Schrodinger

We propose a generative Artificial Intelligence workflow whereby we combine a Generative Adversarial Network (GAN) and Reinforcement Learning (RL) for simultaneous hit identification and optimization.

OpenEye

Here we adopt the hybrid of two different computational strategies for the hit identification of SARS-CoV-2 Nsp3 Mac1 inhibitors.

1. Hit identification via the large-scale virtual screening of Enamine databases

1.1 Initial virtual screening and structural filtering processes

Glide (Schrödinger Release)

Our proposed hit-identification workflow extends pipelines developed by our team during  CACHE Challenge 2. The core methodology consists of high-throughput docking followed by binding affinity estimation using Molecular Mechanics Poisson-Boltzman Surface Area (MMPBSA) on multiple poses drawn from a molecular dynamics (MD) run of the protein-ligand complex.

None

We will use structure-based ultra-large virtual screenings using VirtualFlow. Step 1: Protein preparation Protein structures will be prepared with Maestro from Schrödinger (protonation state assignment, assignment of missing atoms/side chains, hydrogen atoms, ...). MD simulations of the target protein will be carried out using Amber 18. Conformations will be clustered, and representative structures of the clusters will be used for the virtual screens.

Maestro (protein preparation)

CODASS3
This proposal is a substantially enhanced version of our previously successful CACHE2 proposal. It includes improvements to every stage of our COmbined Docking And Similarity Search 2.0 (CODASS2) workflow that was applied to that challenge, as well as introducing additional tools and features to boost both its throughput (and thus the size of its screening library) and the reliability of its predictions. In summary, these improvements are:

None

We developed multi-scale and multi-task neural networks to learn binding structures and binding affinities between compounds and proteins based on our previous works[1-3]. The model takes geometric graph representation of compounds and proteins as input. The compound was processed by a physics-driven graph neural network, integrating the geometry and momentum information into the topological structure.

NA

The hit identification and drug discovery strategy consist in high-throughput docking for the identification of modulators of the NSP3 helicase of SARS-CoV-2.

Schrodinger Drug Discovery suite, BIOVIA Pipeline Pilot, BioSolvIT, MolSoft ICM.

Our approach follows multiple stages that gradually funnel massive ligand libraries into hits,  leads, and optimized leads.  The multiple stages combine earlier data-driven methods and latter principle/physics-driven methods as detailed as follows.  

The Enamine REAL Database (5.5 billion compounds) will be used as target database for a deep learning-accelerated virtual screening campaign against the ADPr site of SARS-CoV-2 Nsp3 macrodomain (Mac1). First, we will remove molecules with a computed Tanimoto index of more than 0.6 from any available Mac1 ligand, in order to prioritize completely novel scaffolds.

Maestro, Glide, ICM

In our workflow we will employ 3D-pharmacophore screening to synergize the information coming from co-crystalized fragment crystal structures with the information from molecular dynamics simulations.

InteLigand - LigandScout

CCG - MOE

Schrodinger - Desmond 

CCDC - GOLD

To identify hit molecules for the macrodomain of SARS-CoV-2 Nsp3, we will use the V-Dock approach developed by our group. The V-dock approach uses deep learning models that predict the protein-ligand docking scores from SMILES strings using the docking results of a subset of the whole library instead of directly docking all ligands. We have already shown that protein-ligand docking scores can be accurately predicted from the SMILES representations.

Glide

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.

in-house

The small molecule libraries will be obtained from the ZINC and Mcule purchasable databases and common filters will be applied to remove duplicates. Additionally, an in-house Evolutionary chemical binding similarity (ECBS) method (PMID: 31504818) will be used for the primary virtual screening of the curated database.

Modular synthon-based approach - V-SYNTHES was published in Nature 601, 452–459 (2022). It first identifies the best scaffold–synthon combinations as seeds suitable for further growth, and then iteratively elaborates these seeds to select complete molecules with the best docking scores.

ICM-Pro is provided by MolSoft.

Using our Binary Star platform, we will employ a comprehensive computational protocol to enable the discovery and optimization of novel lead compounds for the ADPr site of SARS-CoV-2 Nsp3 macrodomain (Mac1). We will run a target analysis workflow (target validation and identification step) for the experimentally determined high-resolution structures of SARS-CoV-2 Nsp3 Mac1, to choose a suitable structure for the virtual screening campaign.

Schrödinger 

AMBER.

We propose to apply a massive library screening workflow that exhaustively screens the 4.5 billion compound Enamine REAL database using a deep-learning-based Drug Target Interaction (DTI) prediction engine to identify molecules likely to bind to SARS-CoV-2 Nsp3.

I have developed a genetic algorithm (GA) that can search Enamine Real Space and will use it to find molecules with good docking scores to the target.

Glide

Our proposed pipeline consists of three steps. As a preliminary step, we will define a binding site around the ADPr site of PDB 7KQB.

Gaussian

Foldit is a crowd-sourced molecular biology game. The CACHE challenge will run as a series of regular puzzles in the Foldit platform and, if prior drug design puzzle experience is any indication, will see participation levels equivalent to other Foldit puzzles.

We will identify the most conserved residues of the Nsp3 Mac1 adenine binding cavity and the proximal ribose site where there are co-crystalized fragments (54 and 9 PBD submissions respectively) and lead-like small molecules (i.e. Gahbauer et al., bioRxiv. 2022). This will be done by performing multiple sequence alignment (MSA) with the Kalign algorithm on approximately 200,000 SARS-CoV-2 Nsp3 sequences from the NCBI.

We will build on an analysis pipeline we have developed that is capable of searching billion-scale small-molecule libraries for binding candidates to a target pocket. In the first phase of the pipeline, we will perform a fast (and approximate) affinity prediction using a strategy based on graph neural networks (GNNs). We have developed GNNs that compute representations of both ligand and protein pocket based on a diverse collection of surface properties.

N/A

Our proposal is to link a set of fragments co-crystallized with nsp3, using either a deep generative model or a knowledge-based linker database, to afford drug-like molecules spanning at least two subpockets of the target. In a first step, 186 co-crystallized fragments (Schuller et al., Sci Adv.

SYBYL x2.1.1, Certara USA Inc., Princeton, U.S.A.

Szybki, Filter: OpenEye Scientific Sofware, Santa Fe, U.S.A.

We will deploy a proprietary deep learning-based framework to rapidly screen multi-billion small molecule libraries. The performance of the proposed framework is tested on several curated as well as publicly-available unbiased benchmarking datasets. To demonstrate the actual application of the framework, we have screened 1.37 billion molecules to discover new inhibitors of the epigenetic protein BRD9 bromodomain.

Maestro (protein preparation)

Glide for docking/hits prioritization (but we'll make a decision later whether to use it or Autodock4/Autpodock Vina or SMINA)