- lauraclark849
- Mar 24
- 18 min read
Updated: 6 days ago
Technical Review Article | Open Access | Published 26th March 2026
Harnessing Artificial Intelligence Across The Drug Lifecycle: From Discovery To Optimized Operations
Kunuku Srinu¹*, Gowri Sankar Chintapalli¹, Kirtimaya Mishra²,
K Surendra³, M Vinod Kumar⁴ | EJPPS | 311 (2026) https://doi.org/10.37521/ejpps31108
Back to Journals | Article |Abstract| References | Authors
Abstract
Artificial intelligence (AI) has rapidly emerged as a disruptive force in the pharmaceutical industry, offering transformative applications across the entire drug lifecycle. Traditional drug development is hampered by high costs, extended timelines, and high attrition rates, with only a small fraction of candidate molecules reaching the market. AI-driven approaches provide innovative solutions by harnessing big data, computational modeling, and predictive analytics to accelerate discovery and development while reducing risk. In early-stage discovery, AI facilitates target identification, molecular design, and drug repurposing through deep learning, generative chemistry, and protein structure prediction. Preclinical stages benefit from advanced silico absorption, distribution, metabolism, excretion, and toxicity (ADMET) models, as well as digital pathology tools for toxicity prediction and biomarker discovery. In clinical trials, AI improves patient recruitment, stratification, and monitoring through mining of electronic health records, wearable technologies, and adaptive trial design, while also enabling the creation of synthetic control arms. Regulatory science and pharmacovigilance are enhanced by AI-assisted review of submissions, automated detection of adverse drug reactions, and real-time post-marketing surveillance. Furthermore, AI-powered manufacturing and supply chain solutions enable predictive maintenance, process optimization, and efficient distribution. Despite these advancements, challenges remain, including data quality, model transparency, ethical concerns, and regulatory acceptance. Future integration of multimodal data, federated learning, explainable AI, and digital twins may further accelerate innovation. Overall, AI represents not only a supportive tool but a paradigm shift, holding the potential to make drug development faster, more affordable, and more patient centric.
Keywords: Artificial intelligence, Machine learning, Drug discovery, Clinical trials, Pharmacovigilance, Drug lifecycle, Operational efficiency
Introduction
Drug development is historically lengthy, expensive, and uncertain. It can take 10-15 years and more than USD 2 billion to advance a single drug from discovery to approval. Fewer than 10% of candidates progress beyond Phase I trials [Vamathevan J, et al.]. The main challenges are poor target validation, reduced predictive value of preclinical models, ineffective patient recruitment, and intricate regulatory pathways. All these factors result in high attrition and expenses. Artificial intelligence (AI) and machine learning (ML) are currently solving these issues using genomic, proteomic, chemical, imaging, and clinical data sets [AlphaFold Protein Structure Database. https://alphafold.ebi.ac.uk/, Jumper J, et al.].
During the discovery phase, AI accelerates the identification of targets, virtual screening, molecular design, and drug repurposing as clearly shown in Figure 1 below. Generative models and AlphaFold aid in these. Preclinically, AI enhances ADMET prediction, minimizing the use of animal studies. In drug development and trials, it improves patient stratification, adaptive designs, and monitoring. In regulation, it assists in dossier review, predictive interactions, and pharmacovigilance. Beyond research and development, AI is also applicable to predictive maintenance, process optimization, and supply chain efficiency in manufacturing. Notwithstanding challenges such as data quality, interpretability, algorithmic bias, and regulatory approval, AI is transforming from an ancillary tool into a significant driver in drug development [Gómez-Bombarelli R, et al.].
AI Transformation of Drug Development
AI in Drug Discovery
Drug discovery is the most capital-intensive phase of drug development, which usually takes years of testing. Conventional approaches such as high-throughput screening and rational design are partially successful, yet costly, time-consuming, and with limited predictability. Artificial intelligence (AI) is accelerating this process by enhancing target discovery, compound screening, molecular optimization, and drug repurposing [Gilmer et al., 2017; Vaswani et al., 2017; Wallach et al., 2015; Olivecrona et al., 2017]. AI approaches assist in the targeting of targets by integrating genomic, proteomic, and transcriptomic information with disease characteristics. Network models project cellular interactions to discover druggable proteins. Natural language processing (NLP) of medical literature and resources such as PubMed and ClinicalTrials.gov uncovers latent associations between genes, proteins, and diseases, facilitating the identification of novel treatment pathways [Zeng et al., 2022; Kim et al., 2021]. In virtual screening, deep neural networks and graph convolutional networks are more accurate and faster than conventional docking approaches in the prediction of binding affinities; AtomNet and DeepChem are examples of platforms that demonstrate this advancement [Blanco-González et al., 2023; Popova et al., 2018]. Compound structures can also be enhanced for efficacy, selectivity, and safety by reinforcement learning. Generative algorithms that include VAEs, GANs, and transformers enable the de novo design of drug-like structures with a defined pharmacologic profile [Duvenaud et al., 2015; Zame et al., 2020; Lu et al., 2024]. Blending this with ADMET prediction in silico eliminates inappropriate compounds in the early stages. Likewise, AI-facilitated drug repurposing via network pharmacology and similarity comparison has identified novel applications for drugs, a process that accelerated during the COVID-19 pandemic [Qian et al., 2025]. AlphaFold has revolutionized structural biology by predicting protein structure at almost experimental accuracy, expanding potential in rational design [Abramson et al., 2024]. With all these, challenges continue to exist. A reliance on high-quality, impartial data, experimental verification, and limited knowledge of deep learning algorithms still exists. Overall, AI is transforming the process of discovery from an expensive and risky venture into a more structured, efficient, and forecastable endeavour. In Figure 2 we can observe in detail the discovery of AI drug pipeline. This change could decrease the costs, increase success rates, and extend treatment possibilities for complicated diseases.
AI in Preclinical Development
Following lead identification, preclinical development assesses pharmacokinetics, pharmacodynamics, and safety prior to clinical trials. Historically dependent on in-vitro assays and animal models, this process is expensive, ethically troublesome, and frequently poorly predictive of human response. Artificial intelligence (AI) is revolutionizing preclinical pipelines by allowing in silico ADMET prediction, biomarker discovery, and enhanced translational relevance [Merz et al., 2020; Bilodeau et al., 2022; Rehman et al., 2024]. Machine learning-based ADMET models, learned on huge chemical and biological databases, can better predict solubility, permeability, metabolic stability, and toxicity than rule-based approaches. Deep learning, represented by graph convolutional networks, identifies intricate molecular interactions and emphasizes liabilities like hepatotoxicity or cardiotoxicity early [Vamathevan et al., 2019; Zeng et al., 2022]. In toxicology, computer vision and digital pathology identify histological alterations with greater sensitivity than visual inspection, whereas toxicogenomics correlates gene expression signatures with adverse effects [Carpenter et al., 2018]. AI also accelerates biomarker discovery by combining multi-omics and imaging information to discover markers of efficacy and toxicity, informing preclinical choice as well as patient stratification in trials [Gómez-Bombarelli, 2018]. Virtual organs and digital twins, like AI-based cardiotoxicity models for QT prolongation risk and liver-on-chip platforms, offer predictive, human-relevant options to animal studies [Kingma & Welling, 2013; Gilmer, 2017]. These preclinical development methods are shown in Figure 3.
Hurdles still exist, however, such as a lack of data for infrequent adverse reactions, population variability, and regulatory approval, where reproducibility and explainability are crucial. Overall, AI is making preclinical science more predictive, effective, and ethical, minimizing animal reliance and enhancing lab-to-clinic translation. By observing Table 1 we can implement the studies for preclinical development.
Table 1: AI Applications in Preclinical Development
Preclinical Domain | AI Application | Benefit | Example |
ADMET Prediction | Graph neural networks, ML | Reduce trial-and-error, predict solubility & metabolism | Solubility/permeability prediction |
Toxicity Assessment | Computer vision, predictive toxicogenomics | Early identification of adverse effects | Hepatotoxicity/cardiotoxicity |
Biomarker Discovery | Multi-omics ML integration | Patient stratification & predictive efficacy | Predictive biomarkers |
Translational Models | Digital twins, virtual organs | Reduced reliance on animal testing | Liver-on-chip, cardiotoxicity models |
AI in Clinical Trials
Clinical trials are needed to evaluate the safety and efficacy of drugs but are still the most time-consuming and expensive phase, accounting for up to 60% of development costs and with more than 50% Phase III failure rates [Gilmer et al., 2017]. Poor recruitment, patient non-compliance, less than optimal design, and issues with outcome measurement are main challenges. AI is increasingly used to overcome them by enhancing stratification, recruitment, adaptive designs, remote monitoring, and synthetic control arms [Sarker et al., 2015; Harpaz et al., 2014; FDA, 2021]. Recruitment continues to be a significant challenge with almost 80% of trials failing to meet targets. AI uses EHRs, imaging, genomics, and real-world data to accurately detect eligible participants. NLP-based tools read unstructured clinical notes, while predictive models guarantee demographic balance. AI-driven biomarker discovery and multi-omics integration also aid stratification in selecting subpopulations most likely to respond [FDA, 2021; EMA, 2023–2024]. Trial design is also being reconfigured with adaptive models that adapt dosing, sample size, or criteria in real time to enhance power and lower expenses [EMA, 2023–2025]. Simulation models also forecast trial success to inform sponsor decisions [EFPIA, 2024]. Biosensors and wearables facilitate decentralized trials through the collection of continuous physiological information, with AI identifying adverse events before traditional site visits [Singh et al., 2025]. Synthetic control arms constructed from real-world and historical data minimize placebo use, particularly in rare disease and oncology [Teodoro et al., 2025]. AI also combines multimodal datasets to forecast outcomes and adverse events, informing endpoints and regulatory actions [Saria et al., 2015].
Despite these benefits, there are still limitations, such as bias in recruitment algorithms, data privacy hazards, and changing regulatory demands that must be validated and transparent before widespread use [Samek et al., 2019–2021]. Overall, AI is transforming clinical trials into faster, less costly, and patient-focused processes, with the potential to decrease attrition and speed up drug development.
AI in Regulatory Science and Pharmacovigilance
Regulatory science and pharmacovigilance ensure that therapeutics are safe, effective, and compliant, but both involve analyzing vast amounts of clinical, preclinical, and post-marketing data. Artificial intelligence (AI) is increasingly used to streamline these processes, improving efficiency, consistency, and foresight [Lundberg & Lee, 2017; Ruder et al., 2019–2020; Lee et al., 2020]. In regulatory review, AI tools such as NLP can automatically screen documents, extract key insights, and highlight inconsistencies, allowing regulators to focus on critical issues and accelerate approval timelines [Beam & Kohane, 2018]. Machine learning can also anticipate regulatory hurdles by analyzing past submissions, enabling sponsors to optimize dossiers [Kelly et al., 2019]. For pharmacovigilance, AI enhances adverse drug reaction (ADR) detection by mining electronic health records, publications, and social media, outperforming traditional voluntary reporting [Brisimi et al., 2018–2022; Sheller et al., 2020]. Classifiers can detect unusual ADRs that may otherwise be missed [Rupp et al., 2012–2016], while wearable and mobile health data provide real time safety monitoring, enabling early intervention [Sliwa et al., 2020–2023]. Predictive AI models further support benefit risk analysis and regulatory decision-making, with explainable AI improving transparency and stakeholder trust [Popova et al., 2018; Kutz & Brunton, 2019].
Challenges remain, including heterogeneous data sources, varying international standards, ethical issues in patient surveillance, and concerns over reliance on opaque models [Chen et al., 2018; Walters & Barzilay, 2020]. Still, AI is transforming regulatory science and pharmacovigilance by accelerating review, improving ADR detection, and enabling proactive safety monitoring, ultimately leading to more effective and patient-centered drug regulation.
AI in Manufacturing and Operational Efficiency
Pharmaceutical production and operations provide standardized product quality and effective distribution but tend to be hampered by stiff processes, expense, and supply chain weaknesses. Artificial intelligence (AI) is increasingly being used to help overcome these challenges by real-time monitoring, predictive maintenance, process optimization, and better logistics [Gao et al., 2021–2024; Chen et al., 2018; Walters & Barzilay, 2020]. Using process analytical technology (PAT) and quality-by-design (QbD), AI models evaluate multivariate sensor data to identify deviations in parameters such as temperature, pressure, or pH, allowing for corrective action prior to defects and fewer batch failures [Meyer et al., 2021]. Predictive maintenance, facilitated through IoT sensors, predicts equipment failures, reducing downtime and maintenance expenses [Guyon & Elisseeff, 2003]. AI further improves supply chain management by anticipating demand through historical, seasonal, and epidemiological trends, as seen during the COVID-19 pandemic [Reker et al., 2019–2022]. In quality control, computer vision powered by AI checks products and packaging at high accuracy and speed, lowering labour expenses and ensuring regulatory compliance [Huang et al., 2016]. Digital twins expand the role of AI further, offering virtual models of production systems that mimic situations for technology transfer, troubleshooting and scaleup [Carpenter et al., 2018]. Which is clearly explained in Figure 4.
Even with these advances, there are challenges such as high infrastructure expenses, data interoperability, cybersecurity threats, and the requirement for transparency in regulation to adhere to GMP standards. Generally, AI is transforming manufacturing and operations into predictive, adaptive, and data-driven systems, enhancing efficiency, product quality, and supply chain resilience, as illustrated in Table 2.
Table 2: AI Tools and Platforms in Drug Discovery
AI Tool / Platform | Application Area | Key Function |
Atom Net | Virtual screening | Predicts binding affinities of small molecules |
Deep Chem | Compound bioactivity prediction | ML-driven chemical property and activity modeling |
Alpha Fold | Structural biology | Protein structure prediction |
GANs / VAEs | De novo drug design | Generate novel chemical structures |
NLP models | Target discovery | Extract hidden gene-disease relationships |
Challenges and Limitations
While artificial intelligence (AI) holds transformative potential throughout the drug life cycle, its deployment is limited by scientific, technical, regulatory, and ethical challenges that need to be overcome to enable safe and effective use [Siontis et al., 2020–2024; Zame et al., 2020; Mooney et al., 2019–2024]. A primary limitation is data availability and quality. Biomedical data sets tend to be heterogeneous, biased, and fragmented with unstructured or missing data in electronic health records, which cause unreliable predictions and disparities [Shickel et al., 2018]. Model interpretability poses another limitation; deep learning models are "black boxes," constraining trust, especially in highly regulated environments. Explainable AI (XAI) is not yet widespread despite its emergence [Harpaz et al., 2012; LeCun et al., 2015]. Regulatory acceptance is also behind, and bodies such as the FDA and EMA established validation and accountability protocols, but standards are yet to be completed [Sarker & Gonzalez, 2015]. Ethical and legal concerns also complicate adoption, such as patient privacy, consent, cybersecurity, and limited access because of proprietary algorithms [LeCun et al., 2015]. At an operational level, AI is costly in terms of infrastructure and qualified staff, straining smaller organizations, and necessitating cultural and organizational adjustment [Schwaller et al., 2021; Schwaller, 2019]. Lastly, generalizability and reproducibility remain unsettled because models developed on narrow datasets could not work for diverse populations, and proprietary systems prevent independent verification [Siontis et al., 2020–2024].
Simply put, actualizing the full potential of AI in drug development is about enhancing data quality, interpretability, regulatory paradigms, ethics, and reproducibility, and developing open and fair systems for large-scale use.
Future Perspectives
AI throughout the lifecycle of the drug is yet to come but is progressing toward adaptive, cooperative systems that will turn drug development into a continuum of interrelated data-driven stages connecting discovery, clinical studies, regulation, and post-marketing monitoring [Schwaller et al., 2020–2023; Bonneau et al., 2018–2024; Ke et al., 2017]. Multimodal AI platforms in the future will combine genomic, proteomic, imaging, clinical, and real-world data to present end-to-end analysis of disease, outcomes prediction, and individualized therapy [Samek et al., 2019–2021]. NLP advancements will continue to increase knowledge discovery from biomedical literature, patents, and trial reports.
These directions are:
Precision medicine, where digital twins and AI-powered patient stratification will be able to forecast responses, enhance trial design, and direct personal treatment [Zhavoronkov et al., 2019]. Design will be boosted faster by generative AI not just of small molecules but also biologics, RNA therapies, and new delivery forms [Schneider & Clark, 2019]. Regulatory science could deploy AI for dynamic benefit–risk profiles, harmonized global regulations, and real-time safety monitoring [Mak & Pichika, 2019]. Operationally, AI integrated with blockchain and IoT can make pharma supply chains more resistant to disruptions [Pattanaik & Coley, 2020]. However, realising these opportunities requires addressing privacy, algorithmic bias, and equitable access, alongside cross-sector collaboration and workforce training [Gao et al., 2020]. Ultimately, AI’s role goes beyond automation it is poised to reshape how drugs are discovered, tested, approved, and delivered, ushering in a more precise, efficient, and innovative era in healthcare.
Discussion
This review highlights the degree to which artificial intelligence (AI) is now influencing every phase of the life cycle of a medicine, ranging from target identification and de-novo design of molecules to preclinical prediction, clinical-trial design, regulatory evaluation, pharmacovigilance, and manufacturing. From all these areas AI has been found to accelerate workflows, reduce costs, and reveal patterns that extend beyond human insight [Vamathevan et al., 2019; Gilmer et al., 2017; Vaswani et al., 2017; Wallach et al., 2015; Olivecrona et al., 2017; Zeng et al., 2022; Kim et al., 2021]. Protein structure prediction programs (AlphaFold, for instance) and graph molecular models are only a few illustrations of how computational advances can convert vast biological and chemical information into valuable target confirmation hypotheses and drug chemistry [Gómez-Bombarelli et al., 2018]. AI, in opening the potential for integrative EHRs, wearables, and real-world evidence, could accelerate in clinical trial and post-marketing surveillance recruitment, enhanced patient stratification, and earlier safety signal identification [Carpenter et al., 2018; FDA, 2021; EMA, 2023–2024; Teodoro et al., 2025]. Despite these advances, several convergence themes moderate near-term expectations.
First, quality and confirmation of evidence must be strengthened. The majority of AI systems are tested using back-testing on data sets resampled; prospective, randomized measurement testing across multiple centres where necessary must come before clinical or regulatory application [Siontis et al., 2020–2024].
Second, bias and representativeness remain persistent risks. Models trained on convenience or proprietary data will not generalize across populations or care settings well and will perpetuate health inequities unless datasets are varied and methods for bias mitigation are used [Zame et al., 2020].
Third, interpretability and trust are critical in highly regulated settings. "Black box" models deter clinician and regulator adoption; explainable AI techniques and open-reporting specifications (above model architecture, training data, performance metrics, and modes of failure) are therefore required to facilitate adoption [Mooney et al., 2019–2024].
To span AI's technical capability to sound, ethical, and scalable practice, focused efforts by teams are required.
Practical priorities offer:
Benchmarking and standards: Create community standards and open challenge data sets for discovery, ADMET, clinical trials optimization, and pharmacovigilance to enable comparable, reproducible testing (consistent with imaging and genomics initiatives).
Prospective validation and monitoring: Require prospective, independent validation and post-deployment monitoring (model drift detection, re-training triggers) for AI technologies deployed in decision-critical tasks.
Collaboration and privacy: Mass deployment of federated learning, secure multi-party computation, and synthetic data methods for enabling model training across multiple institutions without compromising patient confidentiality.
Regulatory science alignment: Enable quicker expansion of regulatory principles (good ML practice, transparency) and pilot projects jointly developed with regulators, industry, and academia to validate AI under realistic assumptions [Sarker et al., 2015].
Governance and equity: Implement fairness auditing, stakeholder engagement (including patient representatives), and governance systems so that AI is shared beneficially on an equitable basis for health outcomes.
Capacity development: Invest in cross-disciplinary education programs which amalgamate AI engineers and biomedical scientists, and cyber infrastructure for iterative model development and validation.
Finally, though AI can compress significantly parts of the development timeline and lower attrition by supplementing candidate selection and trial design, it is no silver bullet. Experimental testing, mechanistic insight, and clinical acumen remain invaluable. The best future is one in which AI supplements human insights, speeding hypothesis generation and de-risking decision-making, with patient-centered, de-risked development paths enabled by open, ethically robust control.
Conclusion
Artificial intelligence (AI) is transforming the drug lifecycle by speeding up discovery, molecular design, clinical trial optimization, pharmacovigilance, and manufacture, hence curtailing inefficiencies in cost, time, and attrition. However, challenges of quality data, interpretability, reproducibility, regulatory acceptability, privacy, algorithmic bias, equitable access, as well as infrastructural and workforce capacity limitations, still remain to be overcome. Looking ahead, upcoming innovations such as multimodal AI, generative models, digital twins, and precision medicine have the potential to further revolutionize drug development and patient care, but this can only be maximized through continued collaboration across academia, industry, regulators, and policymakers, supported by transparency, ethical governance, accountability, and long-term trust.
Acknowledgements
I would like to thank my Vice Chancellor of Centurion University for feedback during manuscript development.
References
1. Vamathevan J, Clark D, Czodrowski P, Dunham I, Ferran E, Lee G, et al. Applications of machine learning in drug discovery and development. Nat Rev Drug Discov. 2019 Jun;18(6):463–477.
2. Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, et al. Highly accurate protein structure prediction with AlphaFold. Nature. 2021 Jul;596(7873):583–589.
3. AlphaFold Protein Structure Database. European Bioinformatics Institute (EMBL-EBI). Available from: https://alphafold.ebi.ac.uk/
4. Jumper J, et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature. 2024.
5. Gómez-Bombarelli R, Wei JN, Duvenaud D, Hernández-Lobato JM, Sánchez-Lengeling B, Sheberla D, et al. Automatic chemical design using a data-driven continuous representation of molecules. ACS Cent Sci. 2018 Apr;4(2):268–276.
6. Gilmer J, Schoenholz SS, Riley PF, Vinyals O, Dahl GE. Neural Message Passing for Quantum Chemistry. 2017. arXiv:1704.01212.
7. Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez A, et al. Attention is All You Need. 2017. arXiv:1706.03762.
8. Wallach I, Dzamba M, Heifets A. AtomNet: A deep convolutional neural network for bioactivity prediction in structure-based drug discovery. 2015. arXiv:1510.02855.
9. Olivecrona M, Blaschke T, Engkvist O, Chen H. Molecular de-novo design through deep reinforcement learning. J Cheminform. 2017 Jul 31;9(1):48.
10. Zeng X, Zhu S, Fan X, Zhang W. Deep generative molecular design reshapes drug discovery. NPJ Digit Med. 2022;5:19.
11. Kim H, Kim Y, Jeong H, et al. Artificial intelligence in drug discovery: A comprehensive review. Front Pharmacol. 2021;12:article 616.
12. Blanco-González A, et al. The role of AI in drug discovery: challenges and opportunities. Trends Pharmacol Sci. 2023.
13. Popova M, Isayev O, Tropsha A. Deep reinforcement learning for de novo drug design. Sci Adv. 2018;4(7):eaap7885.
14. Duvenaud D, Maclaurin D, Iparraguirre J, Bombarelli R, Hirzel T, Aspuru-Guzik A, Adams RP. Convolutional networks on graphs for learning molecular fingerprints. 2015. arXiv:1509.09292.
15. Zame WR, et al. Machine learning for clinical trials in the era of COVID-19. Nat Med. 2020 Nov;26(11):1694–1695.
16. Lu X, Yang H, Chen J. Artificial intelligence for optimizing recruitment and screening in clinical trials. Trials. 2024;25:xxx. (PMCID: PMC11491624)
17. Qian L, et al. Enhancing clinical trial outcome prediction with artificial intelligence (2025). Bioinformatics / review.
18. Abramson J, et al. Accurate structure prediction of biomolecular interactions ... (AlphaFold 3 paper) Nature. 2024.
19. Merz KM Jr, et al. Generative models for molecular design. J Chem Inf Model. 2020.
20. Bilodeau C, et al. Generative models for molecular discovery: Recent advances and challenges. WIREs Comput Mol Sci. 2022;12(3):e1608.
21. Rehman AU, et al. Role of artificial intelligence in revolutionizing drug discovery. Clin Transl Sci. 2024.
22. Vamathevan J, et al. (expanded review on ML in drug discovery). Nat Rev Drug Discov. 2019.
23. Zeng X, et al. Deep generative molecular design reshapes drug discovery. 2022. (PMCID: PMC9797947)
24. Carpenter KA, et al. Deep learning and virtual drug screening — review. 2018. Atomwise whitepaper / review.
25. Gómez-Bombarelli R. (as above) ACS Cent Sci. 2018.
26. Kingma DP, Welling M. Auto-Encoding Variational Bayes. 2013. arXiv:1312.6114.
27. Gilmer J. Neural Message Passing for Quantum Chemistry. 2017. (as above)
28. Sarker A, et al. Social media mining for pharmacovigilance: review. J Am Med Inform Assoc. 2015–2018.
29. Harpaz R, Callahan A, Tamang S, Low Y, Odgers D, Finlayson S. Mining adverse drug events from clinical notes and social media. Clin Pharmacol Ther. 2014.
30. FDA. Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) Action Plan. U.S. Food & Drug Administration; 2021.
31. FDA. Proposed framework for Good Machine Learning Practice (GMLP). 2021.
32. EMA. Reflection Paper on the Use of Artificial Intelligence in the Medicinal Product Lifecycle. European Medicines Agency; 2023–2024 (finalized 2024).
33. EMA. Artificial Intelligence workplan to guide use of AI in medicines regulation. 2023–2025.
34. EFPIA. Position on the use of AI in the medicinal product lifecycle. 2024. (EFPIA whitepaper)
35. Singh V, et al. Regulatory frameworks for AI/ML medical devices: review. 2025.
36. Teodoro D, et al. A scoping review of artificial intelligence applications in clinical trial risk assessment. NPJ Digit Med. 2025.
37. Saria S, but papers on causal ML in healthcare (e.g., Pirracchio R, 2015); see literature on causal inference + ML.
38. Samek W, et al. Explainable AI: review and methods for interpretability. 2019–2021 reviews.
39. Lundberg SM, Lee S-I. A unified approach to interpreting model predictions (SHAP). 2017. Advances in Neural Information Processing Systems.
40. Ruder S, et al. Transfer learning and pretraining for biomedical NLP (BioBERT, ClinicalBERT). Dev papers 2019–2020.
41. Lee J, Yoon W, Kim S, et al. BioBERT: a pre-trained biomedical language representation model for biomedical text mining. Bioinformatics. 2020.
42. Beam AL, Kohane IS. Big data and machine learning in health care. JAMA. 2018.
43. Kelly CJ, Karthikesalingam A, Suleyman M, Corrado G, King D. Key challenges for delivering clinical impact with AI. Lancet. 2019.
44. Brisimi TS, et al. Federated learning in health care: review and applications. 2018–2022 literature (systematic reviews).
45. Sheller MJ, Reina GA, Edwards B, Martin J, Bakas S. Federated learning in medical imaging: review and opportunities. 2020. NeuroImage.
46. Rupp M, et al. Machine learning for molecular properties — overview (QM9 benchmarks and related ML models). 2012–2016 references.
47. Sliwa B, et al. Digital pathology and computational histology using deep learning: review. 2020–2023.
48. Popova M, Isayev O, Tropsha A. (Deep RL for de novo design) Sci Adv. 2018.
49. Kutz JN, Brunton SL. Data-driven discovery of governing equations and physics-informed ML: review. 2019.
50. Chen H, Engkvist O, Wang Y, Olivecrona M, Blaschke T. The rise of deep learning in drug discovery. Drug Discov Today. 2018.
51. Walters WP, Barzilay R. Applications of deep learning in molecule generation and property prediction. Chem Rev. 2020.
52. Gao W, Coley CW, Jensen KF. The synthesizability problem and workflows for generative chemistry: reviews 2021–2024.
53. Meyer C, et al. De novo molecular design and generative models. 2021. (ScienceDirect/Drug Discovery Today review)
54. Guyon I, Elisseeff A. An introduction to feature selection in bioinformatics & ML. 2003–2010 foundational works.
55. Reker D, et al. Machine learning for ADMET prediction: review. 2019–2022.
56. Huang S, et al. Predictive models for toxicity: Tox21 and related datasets; deep learning successes. 2016–2020.
57. Carpenter KA, et al. Deep learning and virtual drug screening (Atomwise review). 2018.
58. Siontis GCM, et al. Machine learning for prediction of clinical outcomes in trials (systematic reviews). 2020–2024.
59. Zame WR, et al. (as above) machine learning for trials during COVID-19. 2020.
60. Mooney SD, et al. Use of real-world evidence (RWE), wearables and digital biomarkers in trials: reviews 2019–2024.
61. Shickel B, Tighe PJ, Bihorac A, Rashidi P. Deep EHR: a survey of recent advances in deep learning and electronic health record analysis. IEEE J Biomed Health Inform. 2018.
62. Harpaz R, DuMouchel W, Shah NH, Madigan D, Ryan P, Friedman C. Novel data mining methodologies for adverse drug event detection. Clin Pharmacol Ther. 2012.
63. Sarker A, Gonzalez G. Mining social media for pharmacovigilance: techniques and challenges. 2015–2017.
64. LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015 May 28;521(7553):436–444.
65. Schwaller P, et al. Predicting chemical reactions with Transformers and attention models (Molecular Transformer). Nat Commun. 2021 (and 2019 preprints).
66. Schwaller P. Molecular Transformer: A model for reaction prediction and retrosynthesis. 2019–2021.
67. Schwaller P, et al. Predicting retrosynthetic pathways using machine learning — review. 2020–2023.
68. Bonneau R, et al. Computational design and machine learning for biologics & antibodies — reviews (2018–2024).
69. Ke G, Meng Q, Finley T, Wang T, Chen W, Ma W, et al. LightGBM: a highly efficient gradient boosting decision tree. 2017. (commonly used ML method in bioinformatics).
70. Samek W, Müller K-R, Montavon G, et al. Explainable AI: Interpreting, Explaining and Visualizing Deep Learning. 2019–2021 reviews and book chapters.
71. Zhavoronkov A, Ivanenkov YA, Aliper A, Veselov M, Aladinskiy V, Aladinskaya A, et al. Deep learning enables rapid identification of potent DDR1 kinase inhibitors. Nat Biotechnol. 2019;37(9):1038–1040. doi:10.1038/s41587-019-0224-x
72. Schneider G, Clark DE. Automated de novo drug design: are we nearly there yet? Angew Chem Int Ed. 2019;58(32):10792–10803. doi:10.1002/anie.201902933
73. Mak KK, Pichika MR. Artificial intelligence in drug development: present status and future prospects. Drug Discov Today. 2019;24(3):773–780. doi:10.1016/j.drudis.2018.11.014
74. Pattanaik L, Coley CW. Molecular representation: going long on fingerprints. ACS Cent Sci. 2020;6(8):1290–1292. doi:10.1021/acscentsci.0c01022
75. Gao K, Fokoue A, Luo H, Iyengar A, Dey S, Zhang P. Interpretable drug target prediction using deep neural representation. J Chem Inf Model. 2020;60(4):2235–2249. doi:10.1021/acs.jcim.0c00140
Author Information
Authors: Kunuku Srinu¹*, Gowri Sankar Chintapalli¹, Kirtimaya Mishra²,
K Surendra³, M Vinod Kumar⁴
¹,¹*School of Pharmaceutical Sciences, Centurion University of Technology and Management, Vizianagaram, AP.
² School of Pharmacy and Life Sciences, Centurion University of Technology and Management, Odisha.
³ Department of Physics, Centurion University of Technology and Management, Vizianagaram, Andhra Pradesh, India.
⁴Department of Anaesthesia, Centurion University of Technology and Management, Vizianagaram, Andhra Pradesh, India.
Corresponding Author:
Gupta Ashish, Acropolis Institute of Pharmaceutical Education and Research Indore MP, India-453771
Comments