Drug discovery is identifying chemical entities with therapeutic potential to safely regulate diseases. This is a time-consuming and expensive process which involves target identification, lead discovery, lead optimization, and preclinical testing. Despite substantial investments, drug discovery has a high failure rate due to the potential drug not showing clinical effectiveness, unexpected toxicities, and challenges in introducing into a competitive market.
Figure 1: The process of drug research and development
However, the use of computational and deep learning approaches has resulted in improved speed, success rate and reduced financial costs in drug discovery. Sample Text
Computational biology in drug design
Initial step of modern drug discovery is target identification, which involves various approaches such as, molecular biology, genomics, proteomics, computational biology, and bioinformatics. Determination of binding sites or active sites on the target protein is important as specific residues within them guide the modification and optimization of the initial lead compound, facilitating the generation of new ligand–target protein interactions.
There could be instances where the engagement of the active site is inadequate due to mutations away from the active site, conformational transitions, drug resistance, and expression levels. Therefore, another critical aspect of drug discovery involves investigating pathogenesis and drug resistance with the use of computational chemistry techniques such as, molecular mechanics, quantum mechanics, and molecular dynamics simulations. These techniques, along with biomacromolecular simulation, effectively reveal the molecular mechanisms of the target protein providing novel insights for drug design.
Computer-Aided Drug Design (CADD)
Computer-aided drug design allows investigation of drug candidates and active molecules having similar biochemical properties, using a broad range of theoretical and computational approaches that are a part of modern drug discovery. Structure-based drug design (SBDD) and ligand-based drug design (LBDD) are considered as important categories of CADD, which have been widely used in lead discovery.
Figure 2: Workflow of structure-based drug design (SBDD) and ligand-based drug design (LBDD)
SBDD starts with target identification, as it relies on the 3D structure of the target and active sites to determine ligand–target interactions. Structures of the target proteins are available in the Protein Data Bank (PDB), whereas some target structures are not obtained yet due to limitations of experimental approaches. In these instances, target structures are predicted according to their sequences with the use of computational approaches such as homology modeling, AlphaFold, and ab initio protein structure prediction.
Figure 3: Prediction of target structures with computational approaches
Next step is the identification of the binding site, which uses the information obtained from site-directed mutation and the co-crystallized complex structures of proteins with ligands. In the absence of prior knowledge, blind blocking is performed, which involves docking across the entire protein surface to discover the most probable binding mode. DeepSite, DoGSiteScorer, COACH, and PocketDepth are some of the tools that predict binding sites using blind blocking.
Then the compounds used for virtual screening are selected from compound libraries such as PubChem, DrugBank, ChEMBL, and ChemDB, and they are filtered based on certain rules, properties, and the synthetic accessibility. For the filtered ligands from libraries, the optimized 3D structure should be modeled.
Each compound from the library is docked into the identified binding site, and the score is evaluated using molecular docking tools such as Autodock, CDOCKER, and SwissDock. Molecular dynamics simulations improve the flexibility of the target protein, obtaining target conformations with well-defined binding cavities, and these simulations can be applied for docking scoring and lead optimization. In lead optimization, ligand–target interactions can be determined using molecular dynamics which guides further development of ligands.
LBDD is applied when target structures are unavailable, but there is information about compounds that display activity against the specific target. LBDD starts with a single compound or a set of compounds among them. Then, based on structural similarities, compounds with physicochemical and structural properties responsible for the given biological activity are identified. Following structure–activity relationships (SARs), suitable analogs are designed improving the properties of the compounds. Pharmacophore modeling and quantitative structure–activity relationship (QSAR) are the commonly used approaches in LBDD.
Artificial Intelligence (AI) in de novo drug design
Artificial intelligence is a recent and promising technique in exploring extensive pharmacological data in drug discovery that has boosted the success rates of drug identification. This involves machine learning, a subfield of AI, and its subfield, deep learning.
Figure 4: Relationship between AI, ML, and DL
De novo drug design (DNDD) involves creating novel chemical entities with computational growth algorithms, allowing the generation of molecules without a starting template. DNDD enables exploration of a broader chemical space and facilitates development of drug candidates optimizing both time and cost.
Figure 5: Overview of the machine learning-based de novo drug design procedure
This process begins with the selection and classification of appropriate data, publicly available data sources, obtaining molecules with desired properties for subsequent model learning. Feature representation methods are then applied to learn and depict molecule structures and properties. Based on the learned representation, the optimal generative model will be selected for de novo molecule generation at the end. Additionally, the generative model is optimized by combining reinforcement learning and property prediction models.
Although computational modeling and AI methods show promise in drug design, challenges persist within the current AI-based framework, which will be refined in the future with the advancements.
Maheshi Weerasekara
3rd Year
References
Zhang, Y., Luo, M., Wu, P., Wu, S., Lee, T., & Bai, C. (2022). Application of computational biology and artificial intelligence in drug design. International Journal of Molecular Sciences, 23(21), 13568. https://doi.org/10.3390/ijms232113568
Zhou, S., & Zhong, W. (2017). Drug Design and Discovery: Principles and applications. Molecules, 22(2), 279. https://doi.org/10.3390/molecules22020279
Prieto‐Martínez, F. D., López-López, E., Juárez-Mercado, K. E., & Medina‐Franco, J. L. (2019). Computational Drug Design Methods—Current and Future Perspectives. In Elsevier eBooks (pp. 19–44). https://doi.org/10.1016/b978-0-12-816125-8.00002-x
Mouchlis, V. D., Afantitis, A., Serra, A., Fratello, M., Papadiamantis, A. G., Aidinis, V., Lynch, I., Greco, D., & Melagraki, G. (2021). Advances in de novo drug design: from conventional to machine learning methods. International Journal of Molecular Sciences, 22(4), 1676. https://doi.org/10.3390/ijms22041676
Image Courtesy
Featured Image:
Figure 1:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9658956/
Figure 2:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9658956/figure/ijms-23-13568-f002/
Figure 3:
https://www.sciencedirect.com/science/article/pii/S235291482200034X
Figure 4:
https://www.researchgate.net/figure/AI-ML-and-DL-relation-Image-reproduced-from-16_fig4_338805782
Figure 5:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9658956/figure/ijms-23-13568-f003/