Artificial Intelligence in Drug Discovery

Traditional drug discovery and development pipelines are lengthy, expensive and prone to errors. Target identification and hit discovery are performed with thousands or millions of compounds, followed by lead optimization, preclinical and clinical studies, potentially spanning over a decade and costing billions of dollars. Despite these investments, the failure rate during clinical trials remains high due to numerous reasons.

Artificial Intelligence (AI) and machine learning (ML) are playing an increasingly important role in drug discovery by speeding up the identification of new therapeutic targets and assisting in drug development. But what are AI and ML exactly? In simple terms, AI refers to technology that allows computer systems to simulate human-like intelligence to perform complex tasks. ML, a branch of AI, involves training algorithms on extensive datasets to detect patterns and make predictions.¹

AI in drug discovery is being applied to address the challenges of traditional methods by accelerating biological data analysis to make meaningful deductions concerning drug efficacy and safety. By predicting drug-target interactions and off-target effects from biological data, AI and ML guide the optimization of drug development pipelines, reducing time, cost and the risk of failures.¹

Foundations of AI in Drug Discovery

Core Concepts and Terminology

While traditional drug discovery relies heavily on trial-and-error experimentation and expert assessment of the results, AI leverages computational models to process large datasets and objectively identify patterns to make predictions. AI-driven drug discovery streamlines biological data mining and can predict novel pharmaceutical interactions from a dearth of literature and experimental results. Furthermore, it enables data-driven in silico simulations of complex biological systems to test the potential effects of drug candidates, which would otherwise take years to test experimentally.¹

Artificial intelligence in drug discovery comprises several machine learning techniques.

• In supervised learning, the computer model is trained on a labeled dataset containing input-output pairs, iteratively adjusting its parameters to improve predictive power. Upon training, the model can predict the outcomes of novel input, such as inferring drug activity from molecular features.²
• During unsupervised learning, the model is used to discover hidden patterns and correlations within a dataset without label-based training. Two types of unsupervised learning are clustering and dimensionality reduction. In clustering, similar data, such as compound structures or patient subtypes, are grouped together. Dimensionality reduction involves reducing the complexity of high-dimensional data while retaining its essential features.³
• Often used in de novo drug design and optimization, reinforcement learning agents make decisions by trial-and-error to maximize cumulative rewards. This method is applied to optimize the design of a drug molecule and to maximize its target interaction and efficacy.⁴

Besides these approaches, deep learning is a specialized form of ML using multi-layered neural networks to model complex relationships. Its main advantage is the ability to handle diverse and high-dimensional data, ranging from chemical structures and genomic sequences to images and clinical records.⁵ Deep learning methods include:

• Artificial Neural Networks: They contain interconnected neurons organized in layers, used to predict drug-target interactions and estimate ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) properties.⁶
• Convolutional Neural Networks (CNNs): As CNNs are developed for image recognition, they are ideal for analyzing histological or radiological images, as well as 3D protein-ligand conformations.⁷
• Recurrent Neural Networks (RNNs): They can handle sequential data such as SMILES strings or gene expression time series.⁸
• Graph Neural Networks (GNNs): In GNNs, chemical structures are depicted by nodes and edges that represent atoms and bonds. They can be used to learn molecular properties and predict efficacy and toxicity.⁹

Applications of AI Across the Drug Discovery Pipeline

The various machine learning tools in AI are applied throughout the drug development pipeline, from target validation to clinical trial optimization.

Early-stage Drug Discovery

AI tools can identify biological targets and potential lead compounds in the early phases of drug discovery. During target identification, AI tools analyze genomics, transcriptomics, proteomics and metabolomics data to make correlations between dysregulated genes or proteins and disease phenotypes.¹⁰ Furthermore, natural language processing (NLP) can be employed to mine literature and databases, such as PubMed and ClinicalTrials.gov, to glean insights into targets and their functional significance.¹¹

Predicting drug-target interactions is crucial during the early stages of drug discovery. Deep neural network tools are now able to predict the likelihood of compounds binding to their targets.¹²,¹³,¹⁴

AI technology is also integrated into computer-aided drug design tools that can assess 3D structures of target proteins to predict docking scores and ligand binding.¹⁵ Conversely, ligand-based drug design platforms assist in creating analogs of known active compounds.¹⁶

Deep learning models guide the design process by predicting physicochemical and pharmacokinetic properties, including solubility, bioavailability and stability. These tools help eliminate compounds that are biologically incompatible despite showing high binding affinities.¹⁷,¹⁸

Another set of tools focuses on Explainable AI (XAI) to make the black boxes in deep networks transparent and interpretable. XAI allows researchers to determine which molecular substructures contribute to activity or toxicity, increasing confidence in AI-assisted decision-making.¹⁹

Compound Screening and Lead Optimization

AI can streamline screening methods and lead optimization by more effectively prioritizing candidates and assisting structural refinement. It enables virtual drug screening platforms to predict hits using previously identified active compounds and 3D structures.²⁰ Additionally, deep learning models integrate traditional docking scores with CNN to boost screening accuracy.²¹

AI can automate and improve the accuracy of compound screening by incorporating large-scale assay data. High-throughput screening, in particular, would benefit from AI to drive automation during image-based phenotypic screening. Compounds with the ideal combination of bioactivity and safety can be identified without relying excessively on empirical lab-based screening. Researchers can improve the efficacy and PK/PD profiles of leads through reinforcement learning and genetic algorithms.²²

De novo Drug Design and Drug Repurposing

AI-based strategies offer the promise of shorter timelines, lower costs, and addressing unmet medical needs, especially for rare diseases and pandemics.²³

AI-based generative models, including autoencoders, generative networks and reinforcement learning, empower de novo drug design to generate novel molecular structures with desirable activity and ADMET properties.⁴

In addition to novel drug design technologies, AI can facilitate the repurposing of existing drugs by mining diverse biomedical data sources to hypothesize novel drug-disease links.²⁴

Preclinical and Clinical Development

Preclinical safety assessment and clinical trial design can be costly and time-consuming, especially due to the increased emphasis on personalized medicine.

AI applications advance in silico safety assessments by detecting potential hepatotoxicity and cardiotoxicity, as well as determining dose thresholds. They improve the simulation outputs of in silico models based on quantitative structure-activity relationships (QSAR) or regression models to reveal metabolic stability.²⁵,²⁶

Supervised, unsupervised and deep learning advance personalized medicine by integrating multi-omics data to streamline patient stratification and predict treatment responses for different subtypes.²⁷ These insights can be supported further by AI-powered digital twins that replicate biological systems and processes in patients, which are used to simulate treatment outcomes.²⁸

Benefits and Impact of AI in Drug Discovery

Machine learning and artificial intelligence can enhance efficiency and timelines in drug discovery by integrating vast datasets, automating laborious processes, and uncovering hidden biological insights. It offers significant advantages across all stages of the drug development pipeline, summarized as follows:¹

• Faster identification of potential drug candidates
• Reduced costs and time in drug development due to early failure prediction
• Improved quality and diversity of hits with novel target-disease and drug-target insights
• Enhanced prediction of drug efficacy, toxicity and ADMET profiles
• The optimization of clinical trial design criteria, such as site selection
• Paradigm shift from generalized to personalized treatment

Challenges and Considerations

Despite the potential benefits, integrating AI into drug discovery is accompanied by significant challenges, including technical limitations, regulatory hurdles and ethical issues.

Data and Technical Challenges

The accuracy and applicability of AI-driven insights rely on the quality of data available. Currently, many public databases contain incomplete and inconsistent data with inadequate annotation and metadata, as well as potential bias due to over-representation of well-studied targets or compounds. These issues can be overcome with standardized data curation and harmonization platforms and by adhering to ontologies and semantic standards. Furthermore, models and code should be stored in open-source repositories to support external validation and collaborations.²⁹

Regulatory and Ethical Considerations

As regulatory frameworks adapt to AI/ML in pharma, the FDA and EMA are concerned about the interpretability and traceability of predictive models, particularly deep learning-based black-box models. More specifically, researchers and pharmaceutical companies might have difficulty interpreting why a model made a prediction, which raises skepticism towards AI-generated drug candidates or targets. These issues are compounded by the risk of breaches or misuse of biomedical data, raising ethical concerns.³⁰ However, efforts to establish transparent XAI models and detailed documentation, with contributions from experts, can ensure that pipelines adhere to legal obligations, such as data governance policies.¹⁹

Adoption and Implementation

A lack of in-house expertise in data science, inadequate computational resources and incomplete datasets can limit a pharmaceutical company’s ability to integrate and implement AI tools. These limitations can especially affect small pharma and biotechnology firms, as well as companies with fragmented IT systems and technologies. Although AI adoption can be lengthy for these companies because of these gaps, the process can be expedited through collaborative initiatives, standardized data management tools and extensive employee training.

Future Directions and Emerging Trends

Although artificial intelligence has been integrated into individual phases of drug development pipelines, constant advancements in data, algorithms and computing power are anticipated to reshape the paradigm of drug discovery. This shift is expected to result in end-to-end automation of AI drug discovery pipelines with precision medicine capabilities.³²

AI-based platforms can help identify or design new drug candidates in months instead of years, drive effective drug repurposing and inform clinical trial design and patient recruitment. Their widespread adoption across academia, industry and regulatory bodies can cement the role of artificial intelligence in drug discovery. The outcome would be most evident in rare disease research and treatment, which is currently limited by data sparsity or a lack of infrastructure for widespread clinical trials. AI solutions can overcome these barriers by mining sparse data, simulating disease pathways and informing the development of therapies.³²

Conclusion

The purpose of AI/ML implementation is not to replace scientists but to complement and amplify their capabilities for hypothesis generation, treatment optimization and decision-making. While the current emphasis is mainly on the accuracy and speed of AI drug discovery, it is equally important to develop transparent, interpretable and regulatory-compliant platforms to combine the strengths of AI technologies and human expertise. Thus, AI-guided pharma could reach its full potential to treat rare diseases, cancer and outbreaks, with customized therapy solutions.

Featured Product

References

1. Dara S, Dhamercherla S, Jadav SS, Babu CM, Ahsan MJ. Machine learning in drug discovery: a review. Artif Intell Rev 2022;55(3):1947-1999.
2. Obaido G, Mienye ID, Egbelowo OF, Emmanuel ID, Ogunleye A, Ogbuokiri B, et al. Supervised machine learning in drug discovery and development: Algorithms, applications, challenges, and prospects. MLWA 2024;17:100576.
3. Polanski J. Unsupervised learning in drug design from self-organization to deep chemistry. Int J Mol Sci 2022;23(5):2797.
4. Atance SR, Diez JV, Engkvist O, Olsson S, Mercado R. De novo drug design using reinforcement learning with graph-based deep generative models. J Chem Inf Model 2022;62(20):4863-4872.
5. Gupta R, Srivastava D, Sahu M, Tiwari S, Ambasta RK, Kumar P. Artificial intelligence to deep learning: machine intelligence approach for drug discovery. Mol Divers 2021;25(3):1315-1360.
6. Xu Y, Liu X, Xia W, Ge J, Ju C-W, Zhang H, et al. Chemxtree: A feature-enhanced graph neural network-neural decision tree framework for admet prediction. J Chem Inf Model 2024;64(22):8440-8452.
7. Vaz JM, Balaji S. Convolutional neural networks (CNNs): concepts and applications in pharmacogenomics. Mol Divers 2021;25(3):1569-1584.
8. Ruiz Puentes P, Valderrama N, González C, Daza L, Muñoz-Camargo C, Cruz JC, et al. PharmaNet: Pharmaceutical discovery with deep recurrent neural networks. PLoS One 2021;16(4):e0241728.
9. Abate C, Decherchi S, Cavalli A. Graph neural networks for conditional de novo drug design. Wiley Interdiscip Rev Comput Mol Sci 2023;13(4):e1651.
10. Ali H. Artificial intelligence in multi-omics data integration: Advancing precision medicine, biomarker discovery and genomic-driven disease interventions. Int J Sci Res Arch 2023;8(1):1012-30.
11. Zirkle J, Han X, Racz R, Samieegohar M, Chaturbedi A, Mann J, et al. Deep learning-enabled natural language processing to identify directional pharmacokinetic drug–drug interactions. BMC Bioinformatics 2023;24(1):413.
12. Öztürk H, Özgür A, Ozkirimli E. DeepDTA: deep drug–target binding affinity prediction. Bioinformatics 2018;34(17):i821-i829.
13. Nguyen T, Le H, Quinn TP, Nguyen T, Le TD, Venkatesh S. GraphDTA: predicting drug–target binding affinity with graph neural networks. Bioinformatics 2021;37(8):1140-1147.
14. Huang K, Fu T, Glass LM, Zitnik M, Xiao C, Sun J. DeepPurpose: a deep learning library for drug–target interaction prediction. Bioinformatics 2020;36(22-23):5545-5547.
15. Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, et al. Highly accurate protein structure prediction with AlphaFold. Nature 2021;596(7873):583-589.
16. Loeffler HH, He J, Tibo A, Janet JP, Voronov A, Mervin LH, et al. Reinvent 4: modern AI–driven generative molecule design. J Cheminform 2024;16(1):20.
17. Korshunova M, Ginsburg B, Tropsha A, Isayev O. OpenChem: a deep learning toolkit for computational chemistry and drug design. J Chem Inf Model 2021;61(1):7-13.
18. Li J, Jiang X. Mol‐BERT: An Effective Molecular Representation with BERT for Molecular Property Prediction. WiCOM 2021;2021(1):7181815.
19. Chandak T, Jayashree J, Vijayashree J. Trends and Advancements of AI and XAI in Drug Discovery. Explainable AI (XAI) for sustainable development: Chapman and Hall/CRC; 2024:233-255.
20. Wu Y, Ma L, Li X, Yang J, Rao X, Hu Y, et al. The role of artificial intelligence in drug screening, drug design, and clinical trials. Front Pharmacol 2024;15:1459954.
21. Zhang X, Shen C, Zhang H, Kang Y, Hsieh C-Y, Hou T. Advancing ligand docking through deep learning: challenges and prospects in virtual screening. Acc Chem Res 2024;57(10):1500-1509.
22. Krentzel D, Shorte SL, Zimmer C. Deep learning in image-based phenotypic drug discovery. Trends Cell Biol 2023;33(7):538-554.
23. Gangwal A, Lavecchia A. AI-driven drug discovery for rare diseases. J Chem Inf Model 2024;65(5):2214-2231.
24. Cortial L, Montero V, Tourlet S, Del Bano J, Blin O. Artificial intelligence in drug repurposing for rare diseases: a mini-review. Front Med 2024;11:1404338.
25. Liu J, Gui Y, Rao J, Sun J, Wang G, Ren Q, et al. In silico off-target profiling for enhanced drug safety assessment. Acta Pharm Sin B 2024;14(7):2927-2941.
26. Smith GF. Artificial intelligence in drug safety and metabolism. Artificial Intelligence in Drug Design: Springer; 2021:483-501.
27. Ghebrehiwet I, Zaki N, Damseh R, Mohamad MS. Revolutionizing personalized medicine with generative AI: a systematic review. Artificial Intelligence Review 2024;57(5):128.
28. Bordukova M, Makarov N, Rodriguez-Esteban R, Schmich F, Menden MP. Generative artificial intelligence empowers digital twins in drug discovery and clinical trials. Expert Opin Drug Discov 2024;19(1):33-42.
29. Tiwari PC, Pal R, Chaudhary MJ, Nath R. Artificial intelligence revolutionizing drug development: Exploring opportunities and challenges. Drug Dev Res 2023;84(8):1652-1663.
30. Nene L, Flepisi BT, Brand SJ, Basson C, Balmith M. Evolution of drug development and regulatory affairs: the demonstrated power of artificial intelligence. Clin Ther 2024;46(8):e6-e14.
31. Kulkov I. The role of artificial intelligence in business transformation: A case of pharmaceutical companies. Technol Soc 2021;66:101629.
32. Huanbutta K, Burapapadh K, Kraisit P, Sriamornsak P, Ganokratanaa T, Suwanpitak K, et al. Artificial intelligence-driven pharmaceutical industry: A paradigm shift in drug discovery, formulation development, manufacturing, quality control, and post-market surveillance. Eur J Pharm Sci 2024;203:106938.