AI-Assisted chemical probe discovery for the understudied Calcium-Calmodulin Dependent Kinase, PNCK
Cite: Essegian, Derek, et al. “AI-assisted chemical probe discovery for the understudied calcium-calmodulin dependent kinase, PNCK.” bioRxiv (2022): 2022–06.
Overview
The study focuses on PNCK, an understudied kinase associated with cancer progression and survival. The researchers employed a three-pronged approach using homology modeling, machine learning, virtual screening, and molecular dynamics to identify small molecules that inhibit PNCK activity. They discovered a hit-series of compounds from commercially available libraries, which can serve as the starting point for future development of PNCK inhibitors. This research aims to accelerate drug discovery and shed light on the function of PNCK in cancer, ultimately leading to the development of a novel drug.
Background
Understudied kinases, such as PNCK, represent a significant gap in our knowledge of the human kinome. The CAMK family, to which PNCK belongs, lacks targeted small molecule inhibitors despite evidence of their involvement in disease progression. Large-scale multi-omics data analysis has identified PNCK as clinically relevant in solid-tumor cancers, particularly kidney cancer. PNCK has been associated with angiogenesis, apoptosis, DNA damage, and cell cycle control pathways. To study PNCK’s biological relevance and validate it as a target, an AI-assisted chemical probe discovery campaign was conducted. Machine learning and homology models were utilized to identify potential inhibitors from a library of over 7 million small molecules for testing against PNCK activity.
Results
The passage describes the process of building and assessing homology models for the protein PNCK and curating compound libraries for ligand-based virtual screening.
The first experiment explains the process of generating homology models for PNCK. Since no crystal structure for PNCK is available, related structures in the CAMK family were used as templates. The amino acid sequence of PNCK was used in a pBLAST search to identify the most homologous structure, which was found to be CAMK1a. Three different structures of CAMK1a were selected as templates for generating the homology models. The models were created using PRIME and underwent refinement and energy minimization. The validity of the models was assessed by removing the co-crystallized ATP from the binding site and then docking it back in using Glide. If the model accurately predicted the pose and binding mode of ATP within a certain range, it was considered valid. The stability of the models was evaluated through molecular dynamics simulations. The models were also compared to predictions made by AlphaFold using structural alignment and RMSD calculations.
The second experiment describes the process of curating compound libraries for ligand-based virtual screening. Three different virtual screening campaigns were conducted. The first campaign used Naïve Bayesian classification models trained on small molecule kinase activity data to predict potential inhibitors of PNCK. The models were trained on data from homologous kinases in the CAMK family. A list of homologous kinase domains was obtained by multiple sequence alignment, and compounds from a commercial library were prioritized based on the predictions from the models.
The second campaign employed a multi-task deep neural network model trained on aggregated data from ChEMBL and the Kinase Knowledge Base. The model made predictions for the entire compound library, and compounds predicted to be active for homologous kinases were prioritized.
The third campaign used 3D shape-based screening. The predicted binding poses of ATP in the PNCK homology models were used to perform shape similarity screening using fastROCS. Compounds from different commercial libraries were screened based on their shape similarity to ATP.
After the virtual screening campaigns, compounds were ranked and selected for further analysis based on their docking scores, clustering, synthetic tractability, and likely binding poses. A total of 64 compounds were selected for chemical screening. The compounds were then subjected to in vitro screening, and the process of docking, molecular dynamics, and screening was repeated until compounds with improved binding were identified.
Overall, these processes involved computational modeling, virtual screening, and compound prioritization to identify potential inhibitors of PNCK for further experimental evaluation.
Discussion
In this project, computational methods were used to aid in the discovery of potential inhibitors for an understudied kinase called PNCK. Since there were no crystal structures or known ligands for PNCK, a homology model of the protein was built using structures of related kinases as templates. A library of compounds was curated for virtual screening, and various filters and machine learning techniques were applied to select a subset of compounds for further analysis. Molecular docking studies were conducted to predict the binding of the selected compounds to PNCK, and the compounds were ranked and clustered based on their docking scores. Manual evaluation and curation of the compounds were performed, considering factors like synthetic tractability and preliminary drug-like properties. Molecular dynamics simulations were also carried out to refine the predicted protein-ligand complexes and study their energetics. The top compounds from each cluster were purchased and tested in binding assays or screens. The results showed that five top scaffolds were identified using different computational methods, and these compounds exhibited inhibitory activity against PNCK. The success of the project can be attributed to the combination of diverse methods and compound libraries. Further research and optimization are needed to develop isoform-specific inhibitors of PNCK with improved potency and drug-like properties. Once a potent chemical probe is obtained, it can be tested in cells and animal models.
FREE ChatGPT Document Q&A: Get questions answered about any document type of any length!
Plug: Please purchase my book ONLY if you have the means to do so, I usually do not advertise, but I am struggling to stay afloat. Imagination Unleashed: Canvas and Color, Visions from the Artificial: Compendium of Digital Art Volume 1 (Artificial Intelligence Draws Art) — Kindle edition by P, Shaxib, A, Bixjesh. Arts & Photography Kindle eBooks @ Amazon.com.
