The advancements in hardware, software, and algorithms have significantly improved drug design and development processes. In general, discovering an effective drug is an extremely complicated process.
Image Credit: metamorworks/Shutterstock.com
Computer-aided drug discovery (CADD) is associated with several theoretical disciplines, such as molecular modeling, chemoinformatics, bioinformatics, and data mining. It is also associated with a wide range of applications of artificial intelligence (AI), particularly machine learning (ML) and deep learning (DL).
These computational processes have significantly reduced the time and expenditure of drug development. Even though the contributions of CADD at different stages of drug discovery are unmistakable, it is imperative to address several challenges and limitations associated with the methods.
Challenges and Limitations of Computational Methods in Drug Discovery
Several methodologies that are used in CADD, such as quantitative structure-activity relationship studies (QSAR), have proved not to be effective. These methods come under the hype cycle with waives of hopes, disappointments, inflated expectations, and effectiveness. Two key reasons for disappointing results are a lack of proper training and poor data reporting.
Even though molecular docking methods are extremely useful, they could generate false expectations due to the misuse of this technique. For instance, molecular docking was used for purposes it was not initially designed to be used for (e.g., this method was not developed for the correlation of docking scores with experimental binding affinities).
Some of the key challenges of the current computational methods in drug discovery have been associated with chemical and biological space, methodology, scientific dissemination, data sharing, and education. These are discussed below:
Chemical and Biological Spaces
Visualization and subsequent analysis of the chemical structures of a compound is a fundamental step of drug discovery. The searchable chemical space enables the identification of many possible compounds quickly. However, continuous exploration of the existing chemical space or libraries of traditional drugs hinders the discovery of new clinically active chemical entities. Identification of novel chemical entities could end up in the formulation of effective drugs against new diseases.
Novel computational approaches are required to identify neglected chemical spaces or detect poorly explored molecules, such as peptidomimetics, peptides, biologics, metallodrugs, and macrocycles, which could be effective against varied diseases. It is essential to recognize the exact regions of the chemical space to identify potential drug components.
The key aspects of drug discovery include identifying and validating molecular targets with therapeutic relevance. Polypharmacological studies help predict interactions of drug candidates with their putative molecular targets. In this context, CADD methods face challenges linked to multi-target property predictions and exploring protein-protein interactions. Better computational methods are required to detect druggable allosteric sites, binding pockets, and transient binding sites to discover novel drugs.
Methods
One of the greatest challenges faced by scientists is to determine the appropriate computational method to design and optimize drug candidates. Since computational chemogenomics continues to be developed and improved, it is important to keep up with the latest developments. It is also essential to select a methodology that helps refine the selection of descriptors to delineate the biological and chemical spaces.
In the context of computational methodology, there is a constant effort to improve methods to boost the hit rates of virtual screening. In addition, there has been a continual effort to improve molecular docking techniques, particularly for the docking of flexible compounds and macrocycles, and covalent and protein-protein docking. There is a need for better molecular modeling of large and complex systems, and prediction of ADMETox-related properties and druggable pockets (e.g., allosteric binding sites).
A major challenge of computational methodology is maintaining high-quality data in public databases. It is of pivotal importance as this data is used to develop models. Some of the newer methodological challenges in drug discovery are driven by AI. For instance, there is a need to improve performance and experimentally validate the results predicted using ML and DL models.
Another challenge is expanding the searchable chemical space via ML-based compound libraries and mining large and ultra-large chemical libraries. Scientists face immense challenges in the selection and rational use of the best AI, DL, and ML methods. Unlike traditional drug discovery processes where speed is a vital issue, computational approaches face challenges in developing personalized phenotypic validation schemes.
Communication and Data Sharing
The lack of communication between researchers hampers the drug discovery process. It is desirable to work synergistically with experts from different science disciplines and collaborate with different research disciplines to improve drug discovery success rates, reduce the chances of duplicating efforts, and avoid the wastage of valuable resources.
Peer review and documentation of scientific findings, along with a proper method of dissemination of information and knowledge, is important. A major challenge is sharing of data and methods to favor transparency and reproducibility. Adequate education and training are required for students and investigators from different disciplines associated with CADD. This will help avoid misapplication and flawed interpretation of the results. In addition, thorough training will prevent false expectations, reduce CADD disappointments, and improve AI interpretability.
The lack of a high-quality dataset for drug repositioning makes it challenging for in silico approaches to evaluate results. Therefore, for computational drug repurposing algorithms, common performance metrics, such as specificity, sensitivity, and precision, are used. The existing computational approaches are mostly one-sided, i.e., drug-centric or disease centric.
Sources
Sadybekov, A.V. and Katritch, V. (2023) Computational approaches streamlining drug discovery. Nature, 616, pp. 673–685. https://doi.org/10.1038/s41586-023-05905-z
Prasad, S. et al. (2022) Present and future challenges in therapeutic designing using computational approaches. Computational Approaches for Novel Therapeutic and Diagnostic Designing to Mitigate SARS-CoV-2 Infection, pp. 489–505. doi: 10.1016/B978-0-323-91172-6.00020-0.
Shah, A. and Jain, M. (2022) Limitations and future challenges of computer-aided drug design methods. Computer Aided Drug Design (CADD): From Ligand-Based Methods to Structure-Based Approaches, pp. 283-297. https://doi.org/10.1016/B978-0-323-90608-1.00006-X
Medina-Franco, L.J (2021) Grand Challenges of Computer-Aided Drug Design: The Road Ahead. Frontiers in Drug Discovery. 1. https://doi.org/10.3389/fddsv.2021.728551
Tautermann, C.S. (2020) Current and Future Challenges in Modern Drug Discovery. In: Heifetz, A. (eds) Quantum Mechanics in Drug Discovery. Methods in Molecular Biology, 2114. Humana, New York, NY. https://doi.org/10.1007/978-1-0716-0282-9_1
Prieto-Martínez, F.D., López-López, E., et al. (2019) Computational Drug Design Methods—Current and Future Perspectives. In Silico Drug Design. DOI:10.1016/B978-0-12-816125-8.00002-X
Further Reading