Challenge #2

Park

Hit Identification
Method type (check all that applies)
Machine learning
Description of your approach (min 200 and max 800 words)

The in-stock 3D molecules from the ZINC20 database or Mcule Purchasable molecules will be subjected to common filters after duplicates are removed and conformers will be generated.

We will apply our proprietary ECBS (Evolutionary chemical binding similarity) method (PMID: 31504818) for primary virtual screening of the curated database. Based on the likelihood that compounds may bind to related targets, ECBS evaluates chemical similarity. Conserved sequences in ligand binding sites can be found in evolutionarily related proteins. Therefore, when a chemical binds to a target, there is a probability that it may also bind to targets that are evolutionarily related.

 The evolutionarily related chemical pairs (ERCPs) and unrelated pairs are distinguished using a binary classifier in the ECBS model, which is based on classification similarity-learning. When the targets of the chemical pairs are the same or share an evolutionary annotation, the chemical pairs are said to be evolutionarily related. The TS-ensECBS model will be utilized for virtual screening (VS) among various ECBS models, because it will be specifically trained to detect chemical pairs binding to a predetermined VS target. The TS-ensECBS model only defines ERCPs from targets that are evolutionarily linked to VS targets, and it integrates multiple ECBS models based on different definitions of evolutionary information about the VS target to reflect a variety of evolutionary information. The TS-ensECBS model assigns each chemical a similarity score between 0 and 1, with a higher similarity representing a higher possibility of binding to the VS target. Our prior work contains the guiding concepts and comprehensive model construction process for the ECBS models (PMID: 31504818).

The highest TS-ensECBS scoring compounds will go through molecular docking with AutoDock VINA and AutoDock. The RNA binding chains from crystal structures in the IDs 7CXM, 5RLH, 5RLZ, 5RML, and 5RMM will be used for docking procedure. After molecules being chosen based on consensus scores from both docking methods, they will be subjected to clustering and finding most common substructure among them. In order to choose a binding pose for end-state binding free energy calculations, the docking poses of shortlisted compounds from both methods will be evaluated with scoring functions to assess protein-ligand interactions. For choosing the final chemicals for experimental validation, pairwise chemical similarity score, visual inspection and binding free energy scores will be considered.

What makes your approach stand out from the community? (<100 words)

The popular QSAR-based ligand similarity method, which ignores data on multiple target-chemical interactions, only employs compounds that bind to a single target protein, although it is used in ECBS. In ECBS, different levels of evolutionary information about targets—such as motif, domain, family, and superfamily—were encoded into molecular binding similarity. The integration of ECBS method for initial screening and the widely used drug discovery approaches for secondary screening can facilitate to identify high affinity NSP13 hits.

Method Name
Evolutionary chemical binding similarity
Commercial software packages used

BIOVIA Discovery Studio Client

Free software packages used

RDKit, AutoDock VINA, AutoDock, graphDelta, Gromacs, gmx_MMPBSA

Relevant publications of previous uses by your group of this software/method

1. Park, Keunwan, et al. "Machine learning-based chemical binding similarity using evolutionary relationships of target genes." Nucleic acids research 47.20 (2019): e128-e128.

2. Lim, Jin Hong, et al. "Drug Discovery Using Evolutionary Similarities in Chemical Binding to Inhibit Patient-Derived Hepatocellular Carcinoma." International journal of molecular sciences 23.14 (2022): 7971.

3. Durai, Prasannavenkatesh, et al. "Evolutionary chemical binding similarity approach integrated with 3D-QSAR method for effective virtual screening." BMC bioinformatics 21.1 (2020): 1-18.

4. Durai, Prasannavenkatesh, et al. "Identification of Tyrosinase Inhibitors and Their Structure-Activity Relationships via Evolutionary Chemical Binding Similarity and Structure-Based Methods." Molecules 26.3 (2021): 566.

Park

Hit Optimization Methods
Method type (check all that applies)
De novo design
Machine learning
Description of your approach (min 200 and max 800 words)

With chemicals found during the hit identification process, the TS-ensECBS method will be updated with the new experimental data to further optimize the model. The in-stock 3D molecules from the ZINC20 database or Mcule Purchasable molecules will then be screened using the most recent TS-ensECBS model, and top-ranked molecules will then go through additional processes that followed previously to identify hits. An alternate strategy uses the REINVENT tool to create new virtual molecules with multiple desired properties based on structural information from hit molecules. REINVENT allows for the attainment of molecules with a range of parameters as a reward by combining a generative model, reinforcement learning, and enhanced scoring function.

The ChEMBL dataset will be used to train a generative model first. The generative model will then be put through transfer learning with the identified hit molecules and used for sampling. To obtain compounds with high molecular docking scores, improved QED scores, and synthesizable qualities, the LibINVENT and DockStream methods will be integrated into REINVENT. The weight for a docking score increases to 2 to emphasize the compatibility of direct receptor-ligand interactions.  Since the hit and output molecules' core structures will be identical, it will be possible to calculate the relative binding free energies for ligand pairs.

What makes your approach stand out from the community? (<100 words)

The TS-ensECBS will be updated with new SAR data derived from hit identification results, which would enhanced its prediction accuracy. By taking into account both the active and inactive compounds from the previous hit identification process, TS-ensECBS combined with traditional methods will provide candidates that can be purchased. However, the focus of the generative modeling method will be to decorate the core structures of previously discovered hits to produce synthesizable compounds with all necessary properties needed to bind to NSP13 while being drug-like.

Method Name
Evolutionary chemical binding similarity
Commercial software packages used

BIOVIA Discovery Studio Client

Free software packages used

REINVENT, LibINVENT, DockStream, Gromacs, pmx, RDKit, AutoDock VINA, AutoDock, graphDelta, gmx_MMPBSA

Relevant publications of previous uses by your group of this software/method

1. Park, Keunwan, et al. "Machine learning-based chemical binding similarity using evolutionary relationships of target genes." Nucleic acids research 47.20 (2019): e128-e128.

2. Lim, Jin Hong, et al. "Drug Discovery Using Evolutionary Similarities in Chemical Binding to Inhibit Patient-Derived Hepatocellular Carcinoma." International journal of molecular sciences 23.14 (2022): 7971.

3. Durai, Prasannavenkatesh, et al. "Evolutionary chemical binding similarity approach integrated with 3D-QSAR method for effective virtual screening." BMC bioinformatics 21.1 (2020): 1-18.

4. Durai, Prasannavenkatesh, et al. "Identification of Tyrosinase Inhibitors and Their Structure-Activity Relationships via Evolutionary Chemical Binding Similarity and Structure-Based Methods." Molecules 26.3 (2021): 566.