The Pragmatic Playbook for AI in Biologics R&D

Artificial intelligence (AI) is now firmly embedded in the language of drug discovery and development, but the narrative surrounding it is shifting. Early claims that algorithms would autonomously design, optimize and deliver new therapeutics have given way to a more nuanced understanding of where AI truly creates value. Rather than replacing traditional R&D pipelines, AI is increasingly seen as (1) a decision-support system that enhances the speed and precision of human-driven processes, including clinical trials, and (2) a high-throughput platform for generating and analyzing probes of biological pathways, for instance using oligonucleotides.

Regulators have adapted to this new reality. The U.S. Food and Drug Administration (FDA) has establishing draft frameworks and guidance for the validation and auditing of AI and machine learning (ML) applications in pharmaceutical development, emphasizing the need for model transparency, traceable data sets, and human oversight throughout the life cycle of a product.¹ This institutional attention reflects how far AI has moved from theoretical hype toward practical integration, even as its limitations remain evident.

Despite a stream of significant investment from both venture capital and global pharma and an overall optimistic view on its transformative potential, most organizations are still struggling to scale AI meaningfully across their R&D functions. Surveys show that more than 60% of pharmaceutical executives now view AI as essential to research efficiency, but fewer than 20% have achieved successful, enterprise-level adoption.² Many promising initiatives remain confined to proof-of-concept projects, hindered by fragmented data systems, uneven data quality, and the absence of shared standards for model validation.

The moment feels familiar. Pharmaceutical R&D has seen comparable hype cycles before, including combinatorial chemistry in the 1990s, which promised to generate entire libraries of new drug scaffolds automatically, and ultra-high-throughput gene–protein expression and screening platforms in the early 2000s, which aimed to map biology at industrial scale. Both technologies advanced scientific understanding but fell short of the sweeping transformation their proponents predicted. AI is proving similar: its potential is real, but its power lies in well-defined, domain-specific applications rather than in wholesale disruption of the discovery process.

By 2025, the industry stands at a pragmatic inflection point. The question is no longer whether AI will change drug development but how to deploy it responsibly, reproducibly, cost-effectively, and in ways that complement human expertise rather than attempt to replace it. The technologies most likely to endure will be those that amplify existing scientific judgment, not those that try to automate it. One thing is certain: AI has vastly accelerated the pace of information retrieval, along with analysis at an organizational level. It has also entered the realm of synthetic reasoning and problem solving. In addition to funding, the number and volume of human inputs, iteratively, into large language models alone produce many research queries and summary deliverables in a small fraction of the time it would have required until just a year or two ago. Because biological, biochemical, and chemical sciences have a legacy of open publication and sourcing, AI is able to build more in these domains without some of the intellectual property constraints in other fields. This powering of retrieval and synthesis can also be used in the intersection of drug science and the biopharmaceutical business, benefitting many functional areas in biopharma companies.

After several years of bold claims (both inside and outside biopharma) and venture-backed experimentation, the flow of capital into AI for drug discovery has become far more selective. The pattern is clear: funding is moving away from “AI-first drug hunters” that sought to discover new biology autonomously and toward AI-enabled decision-intelligence platforms designed to accelerate and de-risk existing programs.^3,4 Investors have grown wary of pure-platform narratives that promise transformative discovery without tangible clinical output. The result is a more disciplined investment landscape, focused on applications that deliver measurable return on investment within two to three development cycles.

This shift is reflected in how both venture capital and Big Pharma are deploying resources. Instead of trying to reinvent the drug discovery process from scratch, major biopharma companies are partnering with technology groups whose AI systems can streamline established workflows. Organizations such as AbbVie have integrated AI to reduce iterative wet-lab cycles and accelerate structure-guided protein optimization, while research institutions like the Wyss Institute have applied AI to repurpose and optimize molecules already known to be pharmacologically active.^5,6 These programs reflect an evolution from speculative innovation to pragmatic acceleration: the value lies not in the novelty of the algorithm but in the efficiency gains it produces across familiar scientific domains.

The same trend can be seen in the growing number of AI–biopharma alliances. Merck’s partnership network, for example, spans collaborations with BenevolentAI, Absci, and Exscientia, each targeting a defined phase of R&D such as target validation or compound optimization rather than end-to-end discovery. Similar agreements between Sanofi and Exscientia, Bayer and Recursion, and AstraZeneca’s expanding AI portfolio, which includes alliances with Algen and others exploring how to leverage CRISPR oligonucleotide libraries to identify new disease targets, show how the industry now views AI as a modular toolset to be integrated into existing pipelines. The underlying rationale is straightforward: investors and strategic partners are seeking near-term productivity improvements, not long-term platform bets that may or may not yield a viable molecule.

In this environment, the companies best positioned to thrive are those that treat AI as an amplifier and accelerator for scientific expertise rather than a substitute for it. Whether deployed to refine lead selection, accelerate structure–activity relationship modeling, or reprioritize indications, AI is becoming an operational tool for better decision-making, not a philosophical redefinition of how drugs are discovered. Funding is following that logic: pragmatic, data-anchored applications are in, and open-ended promises of algorithmic discovery are out.

Across the drug development continuum, key AI applications have matured from theoretical promise to proven practice. The most tangible gains are being realized in areas where AI acts as an analytical partner to experienced scientists: filtering, ranking, and contextualizing information that would otherwise take years to evaluate. These successes are incremental rather than revolutionary, but they are already influencing how drugs are prioritized, developed, and tested. At the basic level of preparing for a new drug discovery project, AI can speed assimilation and distribution of relevant knowledge to the team.

Indication selection has emerged as one of AI’s most practical and underappreciated use cases. Platforms, such as PandaOmics, Causaly, and Lumanity, integrate literature mining, pathway analytics, and clinical outcomes to predict where an existing or investigational molecule is most likely to demonstrate therapeutic benefit.^7–10 By combining diverse data sources — gene expression profiles, molecular interactions, disease ontologies, and patient outcomes — these systems generate ranked indication lists that help developers focus on the most plausible targets first.

This approach has proven particularly valuable for re-prioritizing or rescuing shelved assets. AI can uncover new correlations between biological pathways and disease phenotypes, guiding developers to reapply known compounds where the mechanistic rationale is stronger or patient need more urgent.^3,11 The result is a reduction in early attrition and a shorter timeline to a clinically testable hypothesis. Recent examples include Absci’s ABS-101, an AI-designed anti-TL1A antibody in inflammatory bowel disease, and Avalo Therapeutics’ IL-1β program, a fast-follower approach that leverages predictive modeling to select inflammatory indications most likely to yield rapid proof of concept. Both illustrate how AI-informed prioritization can direct resources toward programs with a higher probability of near-term success.

AI is also transforming the design and execution of clinical trials by identifying the parameters that most affect success. Algorithms trained on historical enrollment and outcome data can refine inclusion and exclusion criteria, optimize endpoint selection, and model statistical power across diverse patient populations.^2,12 Predictive tools can flag sites at risk of poor enrollment, recommend adaptive design elements, and simulate how different cohort compositions might influence trial outcomes.

These capabilities have begun to yield measurable efficiencies. In oncology and rare disease studies, AI-based stratification models have improved recruitment efficiency and enhanced the interpretability of results.¹³ Alector’s Alzheimer’s program, powered to detect a 40% slowing of cognitive decline, and Faron Pharmaceuticals’ Bexmarilimab study, conducted under a single seamless phase II/III protocol aligning with FDA guidance on adaptive design, illustrate how data-driven design can accelerate timelines while maintaining regulatory rigor. The agency’s evolving stance underscores that AI-generated insights are welcome, but only when paired with transparent methodologies and human oversight.¹

The integration of AI into biomarker identification and patient selection has become indispensable, particularly for biologics and immunomodulatory therapies where patient heterogeneity can obscure efficacy signals. By combining multi-omic data sets with imaging and clinical records, AI models can identify subtle patterns that distinguish responders from non-responders.^14,15 This capacity to define molecular subtypes not only improves trial design but can also inform dosing, combination strategies, and real-world treatment optimization.

Success increasingly depends on matching therapies to specific biological contexts rather than pursuing novelty for its own sake. Drugs targeting similar pathways may yield dramatically different results, depending on the biomarker-defined population. As a result, One of AI’s greatest values lies not in predicting which molecules will work universally but in clarifying the subset of patients most likely to benefit, in a manner analogous to its role in indication prioritization.

Machine Learning (ML) models have achieved notable success in predicting structure–function relationships that guide protein engineering. These algorithms evaluate how minor sequence changes influence binding affinity, stability, and immunogenic potential.^14,16 Applied to antibodies, bispecifics, and Fc-fusion constructs, AI accelerates the cycle of hypothesis generation and experimental validation. Internal programs at major firms have shown that these approaches can reduce the number of wet-lab iterations required for optimization.⁵

Examples across the industry illustrate how AI-guided structural refinement translates into clinical promise. Aclaris is advancing ATI-052, a bispecific anti-TSLP/IL-4R antibody designed through data-driven epitope mapping; RAPT is developing RPT904, an anti-IgE therapy optimized for improved pharmacokinetics and dosing flexibility; and Adagene’s ADG126 incorporates modeling-based insights to guide triplet chemo combinations. Each demonstrates that AI’s most immediate impact lies in improvement, not invention, fine-tuning molecular precision within known therapeutic classes.

AI is beginning to influence formulation design, predicting manufacturability, solubility, and delivery compatibility, domains traditionally dominated by empirical trial and error.¹⁷ For biologics and nucleic acid-based therapeutics, where physical properties dictate stability and bioavailability, predictive modeling helps narrow formulation candidates before lab testing.

Two examples underscore the range of possibilities. Oculis’s OCS-02 program applies computational modeling insights to explore converting subcutaneous delivery into a topical ocular route, addressing formulation barriers that once made such transitions unlikely. Inovio’s DNA-encoded monoclonal antibody (DMAb) platform, meanwhile, uses AI-guided optimization to minimize immunogenic sequences and optimize in vivo expression kinetics. These developments highlight a growing trend toward the integration of computational insight throughout the manufacturing and delivery continuum.

Taken together, these examples represent AI’s quiet revolution in drug development. Rather than disrupting the fundamentals of biology, AI is reinforcing them by providing the analytical scaffolding that allows developers to act faster, design smarter, and fail less often.

Beyond today’s proven use cases, a range of more ambitious applications of AI in drug discovery continue to attract attention and investment. These areas hold genuine promise but remain scientifically or operationally immature, limited by data quality, biological complexity, and the challenge of validation in vivo. While progress in each is real, expectations have thus far outpaced demonstrated capability.

AI models have grown increasingly sophisticated in their ability to map and predict biological network perturbations: how gene expression, protein interactions, and signaling cascades shift in response to a given intervention. However, the leap from correlation to causation remains elusive. Although modern graph-based and transformer models can generate plausible pathway maps, they often lack interpretability and mechanistic grounding.^18,19 The difficulty lies not in modeling the networks themselves but in distinguishing whether an observed change is a downstream effect, an epiphenomenon, or a causal driver of disease.

Progress in this field is constrained by data noise and inconsistency. Multi-omic data sets are often assembled from disparate experimental conditions, using varied annotation standards and incompatible normalization methods. This heterogeneity impairs the reliability of AI predictions and makes cross-study validation difficult. Until data curation improves and causal inference models can be paired with experimentally testable hypotheses, these approaches will remain exploratory rather than decisive.

Predicting immune reactions to biologics remains one of the most complex frontiers in drug development. AI models that analyze peptide sequences and structural motifs are beginning to identify regions likely to elicit immune responses, potentially reducing the risk of formation of anti-drug antibodies.¹⁶ However, immune system behavior is inherently stochastic and context-dependent, shaped by genetic background, comorbidities, and even microbiome composition. Training data sets for immunogenicity modeling remain too limited to capture this variability comprehensively, although that may change as more data become available.

At best, current algorithms serve as early warning systems, flagging potentially immunogenic regions before laboratory testing. They can support design-stage decision-making, such as the sequence refinement approaches now seen in advanced antibody and fusion-protein engineering, but they cannot yet replace empirical immunoassays or human judgment. The field’s maturation will depend on integrating AI with longitudinal clinical safety data rather than relying solely on static molecular features. Progress in this area will benefit from new molecular markers and probes related to tracking immunogenicity.

Another frontier area involves predicting how complex biologics distribute throughout the body and, for neurological indications, whether they can cross the blood–brain barrier (BBB). Generative models have shown potential in designing constructs with properties associated with central nervous system (CNS) penetration, such as molecular weight and surface charge patterns,¹⁶ but at present even the most advanced simulations fall short of accurately predicting in vivo pharmacokinetics.

Differences between preclinical models and human physiology remain a persistent gap, especially in CNS drug development. Programs such as Janssen’s anti-tau antibody, designed to reach midbrain targets, exemplify the challenge: despite informed molecular engineering, true BBB permeability must still be confirmed empirically. AI can narrow the search space, but translation from computational predictions to clinical efficacy remains unreliable. Until in vivo correlation improves, these models will serve primarily as hypothesis generators rather than decision-making tools.

Among the most heavily discussed potential AI applications reflects the possibility of designing entirely new chemical entities or identifying novel biological targets without prior human direction. While generative chemistry algorithms can produce millions of theoretically viable molecules, the overwhelming majority are chemically implausible, synthetically inaccessible, or biologically irrelevant. In practice, medicinal chemists still perform extensive triage, filtering AI-generated compounds through established structure–activity heuristics and known pharmacophore models.^11,19

The same applies to AI-driven target discovery, where pattern recognition from omics or text-mined data may highlight associations but rarely yields actionable mechanisms. Despite claims of “AI-discovered” drugs, nearly all molecules entering clinical testing are scaffold optimizations of known actives or analogues of validated classes rather than breakthroughs in first-in-class biology. These limitations are most evident in large, heterogeneous disease populations, such as metabolic disorders, inflammatory conditions, or oncology subsets, where biological variability defies algorithmic simplification.

For now, generative chemistry and de novo target identification remain valuable for expanding chemical space and hypothesis generation but are not yet capable of autonomous discovery. AI continues to assist the explorer rather than act as the explorer itself.

Amid the measurable progress that AI has brought to modern drug development, some of the most widely circulated narratives still outpace the technology’s demonstrated reality. The hype surrounding AI’s capabilities can obscure its genuine strengths by creating expectations it cannot yet meet. Understanding what remains overstated is essential, not to diminish innovation but to focus investment and scientific energy where it can produce sustainable returns.

As discussed above, the idea that algorithms can independently design and validate new therapeutics persists in popular discussion, but every publicly disclosed example of an “AI-discovered” molecule to date has depended heavily on expert curation and laboratory confirmation.^14,19 While AI can generate molecular structures or predict binding affinities, it lacks the biological context required to determine whether those molecules will be safe, efficacious, or manufacturable. Even in programs positioned as “AI-first,” such as antibody discovery initiatives, scientists still define the target, interpret the predictions, and design the experimental assays that determine whether any computationally generated candidate has practical potential. The automation of early design steps accelerates iteration, but it does not replace the interpretive loop that makes discovery meaningful.

AI systems are only as reliable as the data they learn from, yet much of the biomedical data ecosystem remains fragmented, inconsistently labeled, or biased toward well-studied conditions. Large volumes of poorly annotated information can amplify noise rather than insight, leading to overfitted models and misleading correlations.^3,18 The challenge is compounded by publication bias, which overrepresent positive results, and by proprietary data silos that limit model generalization. As a result, scaling AI does not guarantee better performance. The most successful efforts rely on carefully curated data sets, standard ontologies, and continuous human review to ensure that algorithms capture causality rather than coincidence.

Automation has improved throughput and reproducibility in certain parts of the discovery process, but biological research, regulatory evaluation, and manufacturing still require human interpretation and ethical judgment. Regulatory authorities have made clear that explainability and validation remain prerequisites for any AI-derived submission. The FDA’s current guidance on AI and machine learning emphasizes traceability and “human-in-the-loop” oversight, underscoring that responsibility for decision-making cannot be delegated to a black box.^1,11 Similar limitations apply downstream: process development, clinical evaluation, and quality assurance depend on tacit knowledge, contextual reasoning, and adaptability, qualities that no algorithm currently replicates.

Ultimately, AI is powerful because it helps experts ask better questions, not because it can answer them independently. The technology’s value lies in amplifying human intuition and enabling evidence-based decision-making. Recognizing its boundaries is not pessimism; it is the prerequisite for deploying it responsibly and effectively across the biopharmaceutical landscape.

The most productive applications of AI in drug discovery depend not on the sophistication of the algorithm but on the expertise of the people interpreting it. Successful development teams increasingly recognize that AI’s purpose is to augment rather than replace scientific judgment. Human researchers remain essential for contextualizing uncertainty, identifying false positives, and designing confirmatory experiments that transform algorithmic predictions into validated results. It is still human scientists who can use their own knowledge, reasoning, and intuition to bring data points from different but related field together in moments of insight to solve a previously intractable problem in drug design and development.

This hybrid approach where data-driven insights guide experienced decision-making has proven to be the most sustainable model for innovation. In practice, AI provides a ranked list of possibilities, while human teams determine which of those possibilities are plausible, ethical, and economically viable to pursue. Fab Biopharma’s model exemplifies this philosophy: AI is used to prioritize targets or indications rather than dictate them, and expert scientists apply mechanistic reasoning to convert predictions into testable hypotheses. This process accelerates discovery without relinquishing control over the essential interpretive steps that ensure rigor.

The growing consensus across the industry reflects this same logic. Studies and industry reports have found that “hybrid intelligence” — the intentional coupling of human and machine expertise — consistently outperforms fully automated (or fully human!) systems in both accuracy and reproducibility.^6,11 These findings reinforce that the strength of AI lies in amplifying human capability, not in supplanting it. Teams that treat AI as a partner rather than an oracle are better equipped to recognize data limitations, identify novel opportunities, and integrate cross-disciplinary knowledge into their decision-making. The greater the iteration and connectedness between humans and AI, the better both become.

Despite advances in modeling and automation, AI remains unable to synthesize the full spectrum of considerations that define real-world drug development. No system can yet account for the interplay among mechanistic pharmacology, clinical strategy, manufacturing feasibility, intellectual property positioning, and commercial viability. These domains require trade-offs that depend on judgment, experience, and an understanding of human behavior, all attributes that resist quantification.

In that sense, AI’s most important contribution is not to replace experience but to make it more actionable. By giving experts faster access to patterns, evidence, and predictive insights, AI strengthens the link between data and decision. The future of drug discovery will belong to organizations that cultivate this synergy, combining computational precision with the intuition that only years of experimentation can teach.

As AI becomes more deeply embedded in drug discovery and development, its governance has emerged as a defining issue. Regulatory agencies and industry leaders are converging on the principle that trust, transparency, and traceability must underpin every AI-enabled process. The FDA and the European Medicines Agency (EMA) now emphasize explainability, auditability, and data provenance as prerequisites for use in regulated environments.^1,12 These requirements ensure that any AI-derived decision can be reconstructed, validated, and defended under regulatory scrutiny, a critical safeguard in a field where patient outcomes and public confidence are at stake.

Establishing explainability also helps address one of the central ethical challenges in AI adoption: bias. Algorithms trained on unbalanced datasets risk perpetuating inequities in drug development by favoring well-studied populations or disease areas. Building systems that can both detect and mitigate bias is essential for trustworthy adoption.¹⁵ The focus is shifting from model performance alone to the provenance and representativeness of the data that underlie it. This evolution mirrors the industry’s broader shift toward equity in clinical research and personalized medicine, where inclusion and data integrity are inseparable from innovation.

Operationally, companies are learning that the highest return on AI investment comes not from late-stage automation but from integrating computational tools early in the research process while maintaining explicit human checkpoints. Analyses from McKinsey, Holzinger, and Niazi converge on this finding: embedding AI at the design and prioritization stages delivers the most value, provided decision-making authority remains with expert teams.^2,3,10 While some jobs may be reduced or lost to the combination of AI and robotics, others will be created to use those outputs. It is critical that we constantly train scientists to adapt to this changing environment.

The next challenge lies in scaling these systems responsibly. As global collaborations and multi-omic data sets expand, protecting intellectual property and patient privacy becomes more complex. Secure data-sharing frameworks will be necessary to enable cross-institutional learning without compromising confidentiality. Achieving interoperability across diverse data sets — academic, commercial, and clinical — will be key to realizing AI’s full potential at scale.

Ultimately, even as AI grows more capable, the final review of a therapeutic window balancing safety and efficacy must remain a human responsibility. Machines can model probabilities, but they cannot assign values to risk, compassion, or clinical judgment. Building responsible and scalable AI in biopharma therefore requires a balance between innovation and accountability: systems transparent enough to be trusted, flexible enough to evolve, and humble enough to keep the human in command.

For sponsors and biotech leaders navigating the rapidly maturing landscape of AI in drug development, the question is no longer whether to adopt AI but where and how to deploy it strategically. The most successful adopters are focusing their investments on areas that produce measurable improvements in decision-making while avoiding the temptation to chase unproven or opaque technologies. A pragmatic approach emphasizes targeted integration, human oversight, and return on insight rather than sheer automation.

Organizations can invest confidently in applications that have already demonstrated clear value. Indication prioritization, patient stratification, and trial-powering algorithms consistently reduce time and cost while improving the probability of clinical success. These tools leverage historical data and mechanistic understanding to direct resources toward the most promising therapeutic opportunities, enabling smarter pipeline management. Likewise, structure–function prediction for biologics and antibodies has matured into a reliable contributor to molecular optimization, improving binding affinity, manufacturability, and safety profiles. In both domains, AI acts as an accelerator, making strong scientific intuition operational at scale.

Other areas warrant close observation but cautious investment. Generative de novo target discovery, for example, remains a largely experimental pursuit. It can inspire creativity but has yet to deliver reproducible breakthroughs without human curation. Network biology, safety prediction, and biodistribution modeling are also advancing, but they continue to face data-quality limitations and incomplete biological validation. Companies exploring these frontiers should maintain rigorous performance benchmarks, demand interpretability, and resist the urge to deploy algorithms faster than their evidentiary base supports.

Finally, sponsors should avoid overreliance on systems that function as “black boxes.” Models that cannot explain their reasoning or withstand regulatory examination will offer little long-term value, regardless of how sophisticated they appear. Transparency, reproducibility, and traceable logic remain the currencies of credibility in regulated science. The most effective AI frameworks are those designed to be audited, challenged, and improved over time.

As the biopharmaceutical sector continues to integrate computational intelligence into its workflows, one guiding principle stands out: the best use of AI is to make the right bets faster—not to pretend biology has been solved. The goal is not to surrender decision-making to algorithms but to ensure that every human decision is better informed. By combining the predictive power of AI with the discernment of experienced scientists, organizations can achieve the most important outcome of all: progress that is both accelerated and accountable.

1. “Artificial Intelligence for Drug Development.” U.S. Food and Drug Administration. Accessed 13 Oct. 2025.

2. “How artificial intelligence can power clinical development.” McKinsey & Company. 22 Nov. 2023.

3. Niazi, Sarfaraz K and Zamara Mariam. “Artificial intelligence in drug development: reshaping the therapeutic landscape.” Ther. Adv. Drug Saf. 16: 20420986251321704 (2025)

4. Ogorek, Benjamin, et al. “AI-Powered Drug Classification and Indication Mapping for Pharmacoepidemiologic Studies: Prompt Development and Validation.” JMIR AI. 4: e65481 (2025).

5. “Three ways AI is changing drug discovery at AbbVie.” AbbVie. Accessed 13 Oct. 2025.

6. Alucozai, Milad, Will Fondrie, and Megan Sperry. “From Data to Drugs: The Role of Artificial Intelligence in Drug Discovery.” Wyss Institute. 9 Jan. 2025

7. “Indication Prioritization.” PandaOmics. Accessed 13 Oct. 2025.

8. “Target ID & Prioritization.” Causaly. Accessed 13 Oct. 2025.

9. “Indication Prioritization and Early Clinical Development Roadmap for a Broadly Applicable T-cell Modulatory Therapy.” Lumanity. 30 Apr. 2024.

10. TPP Development & Indication Prioritization. Erik Holzinger Group. Oct. 2023.

11. Sun, Duxin and Christian Macedonia. “Will AI revolutionize drug development? Researchers explain why it depends on how it’s used.” JHEOR. 7 Jan. 2025

12. Yan, Chao, et al. “Leveraging generative AI to prioritize drug repurposing candidates for Alzheimer’s disease with real-world clinical validation.” npj Digital Medicine. 7: 46 (2024).

13. Ocana, Alberto, et al. “Integrating artificial intelligence in drug discovery and early drug development: a transformative approach.” Biomarker Research. 13: 45 (2025).

14. Zhang, Kang, et al. “Artificial intelligence in drug development.” Nature Medicine. 31: 45–59 (2025).

15. Blanco-Gonzalez, Alexandre, et al. “The Role of AI in Drug Discovery: Challenges, Opportunities, and Strategies.” Pharmaceuticals (Basel). 16: 891 (2023).

16. Fu, Chen and Qiuchen Chen. “The future of pharmaceuticals: Artificial intelligence in drug discovery and development.” Journal of Pharmaceutical Analysis. 15: 101248 (2025).

17. Brazil, Rachel. “How AI is transforming drug discovery.” Pharmaceutical Journal. 3 Jul. 2024

18. Gold, E Richard and Robert Cook-Deegan. “AI drug development’s data problem.” Science. 388: 131 (2025).

19. King, Anthony. “Four ways to power-up AI for drug discovery.” Nature. 27 Feb. 2025.