GUEST COMMENTARY—For millions of patients, the slow pace of drug discovery means years – sometimes decades – without effective treatments. Many diseases remain untreatable, not because they are rare or unknown, but because their biological targets are so difficult to drug.
Drug discovery, however, is fraught with challenges, with notoriously high attrition rates – roughly 90% of drug candidates fail to reach approval in clinical trials.1 Many setbacks stem from targets deemed “undruggable”; these are often proteins with flexible, shape-shifting structures or shallow surfaces that lack the stable binding pockets needed for conventional small-molecule drugs.2 Ion channels and G-protein coupled receptors (GPCRs) represent two vitally important but challenging drug targets due to their complex molecular structure and hard-to-reach binding. Ion channels are different from GPCR’s functioning as membrane pores that directly facilitate ion flow in response to stimuli, enabling rapid electrical signalling.
Despite their established involvement in many diseases, an estimated 227 GPCRs remain undrugged. These challenges with drug targeting drive up research and development costs – often exceeding $1-2 billion and over a decade of work per successful new medicine3 – and leave patients without effective treatments for diseases whose key molecular drivers remain unaddressed.
Generative artificial intelligence (GenAI) has emerged as a promising tool to help tackle these obstacles in novel ways (see below). However, enthusiasm must be tempered: AI-generated solutions can be inaccurate. AI hallucinations are incorrect or false responses generated by artificial intelligence models, particularly large language models (LLMs) and GenAI systems. This is typically caused by errors or limitations in the original datasets leading to inaccurate outputs.
High-quality data is paramount in overcoming this. Without robust and task-relevant datasets, even the most advanced models are at risk of producing unreliable outputs. Ensuring data integrity and validation is, therefore, critical to translating AI-driven insights into safe and effective therapies.4
GenAI is reshaping how drug discovery is approached on a broader scale. Instead of relying solely on traditional high-throughput, trial-and-error screening or structure-based design, AI-driven models can generate and refine more efficacious molecular candidates with unprecedented speed and precision. By analyzing vast datasets, GenAI predicts binding interactions, optimizes key pharmacokinetic properties, and helps navigate the complex landscape of drug safety. This shift is enabling entirely new strategies for tackling diseases without suitable treatment options.
Key applications of GenAI in drug discovery include:
Beyond molecule design, GenAI can improve efficiency and cost-effectiveness by filtering out weak candidates early in development. Instead of advancing less effective molecules through costly laboratory testing, AI models can predict potential failures before they reach preclinical stages. This data-driven triage reduces attrition rates and shortens drug discovery timelines.10
While there have not been any de novo AI-designed drugs that have cleared Phase III trials and approved for use so far, AI is becoming increasingly adopted within drug discovery and is shaping industry practices. AI tools are accelerating compound selection, optimizing drug properties, and repurposing existing therapies, bringing a shift from theoretical ideas to real-world applications.
Companies already succeeding in using AI in drug discovery (see Table 1) include Healx, which recently dosed the first patient in a Phase 2 trial with HLX-1502, an AI-discovered therapy for rare tumour disorder neurofibromatosis type 1. And BenevolentAI used its proprietary AI platform to identify baricitinib, originally approved for rheumatoid arthritis, as a potential treatment for COVID-19.
Table 1: A selection of active AI drug discovery programs.
AI is also being applied to traditionally hard-to-drug targets, particularly through pharmaceutical partnerships with smaller, specialized AI-driven biotech firms to help accelerate drug discovery. In the past six months alone, there have been several pharma-biotech partnerships, including Nxera Pharma, who has partnered with my company, Cardiff-based techbio company Antiverse, to design novel GPCR-targeted antibody therapeutics using GenAI. Another example is Insilico Medicine, which has partnered with Tenacia Biotechnology in a project focusing on using generative AI for the early discovery stage of novel CNS disease therapies, developing small-molecule inhibitors from scratch.
If these AI-led methods continue to prove effective, the impact could be far-reaching. Generative models may unlock new treatments for major diseases such as cancer, heart disease, and neurodegeneration, where conventional therapies fall short.
While GenAI has the potential to reshape drug discovery, its success depends on the quality and availability of data. AI models rely on vast, diverse datasets to generate meaningful predictions, but data constraints pose a major challenge. Many biotech datasets are locked in data silos, forcing researchers to duplicate existing work instead of building upon it. This inefficiency slows progress and hinders the potential impact of AI.
Even when data is publicly available, quality and reliability are not guaranteed. Biases, errors, and incomplete reporting are common, particularly in cases where positive results are prioritized over replication studies. In drug discovery, where AI models must distinguish subtle molecular interactions, these inaccuracies can lead to misleading predictions.11
Without high-quality, biologically representative datasets, GenAI’s predictive power is limited, increasing the risk of false positives and ineffective drug candidates.12 As a result, rigorous in vitro and in vivo validation remains essential to confirm that AI-designed therapies not only work as predicted, but also produce meaningful therapeutic effects and lack harmful side effects.
GenAI offers a realistic path forward in tackling previously undruggable targets, with the potential to reshape the entire therapeutic landscape and the whole drug discovery pipeline. By designing novel molecules, optimizing key attributes, and identifying strong candidates earlier, GenAI is accelerating drug discovery and also expanding what is scientifically possible.
However, the success of AI-driven drug discovery depends on more than just technological advancements. Data integrity must come first – before AI-generated molecules can advance through experimental validation and demonstrate real-world efficacy. Achieving this will require close collaboration between data scientists, medicinal chemists, biologists, and regulatory experts, ensuring that AI-driven insights translate into clinically meaningful breakthroughs, offering renewed hope to patients still waiting for effective treatments.
References
Murat Tunaboylu ([email protected]) is the co-founder and CEO of Antiverse, a start-up working on computational antibody design technology based in Cardiff, Wales.
The post How AI is Cracking Medicine’s Most Challenging Drug Targets appeared first on GEN - Genetic Engineering and Biotechnology News.
Drug discovery, however, is fraught with challenges, with notoriously high attrition rates – roughly 90% of drug candidates fail to reach approval in clinical trials.1 Many setbacks stem from targets deemed “undruggable”; these are often proteins with flexible, shape-shifting structures or shallow surfaces that lack the stable binding pockets needed for conventional small-molecule drugs.2 Ion channels and G-protein coupled receptors (GPCRs) represent two vitally important but challenging drug targets due to their complex molecular structure and hard-to-reach binding. Ion channels are different from GPCR’s functioning as membrane pores that directly facilitate ion flow in response to stimuli, enabling rapid electrical signalling.
Despite their established involvement in many diseases, an estimated 227 GPCRs remain undrugged. These challenges with drug targeting drive up research and development costs – often exceeding $1-2 billion and over a decade of work per successful new medicine3 – and leave patients without effective treatments for diseases whose key molecular drivers remain unaddressed.
Generative artificial intelligence (GenAI) has emerged as a promising tool to help tackle these obstacles in novel ways (see below). However, enthusiasm must be tempered: AI-generated solutions can be inaccurate. AI hallucinations are incorrect or false responses generated by artificial intelligence models, particularly large language models (LLMs) and GenAI systems. This is typically caused by errors or limitations in the original datasets leading to inaccurate outputs.
High-quality data is paramount in overcoming this. Without robust and task-relevant datasets, even the most advanced models are at risk of producing unreliable outputs. Ensuring data integrity and validation is, therefore, critical to translating AI-driven insights into safe and effective therapies.4
GenAI: a new approach to drug discovery
GenAI is reshaping how drug discovery is approached on a broader scale. Instead of relying solely on traditional high-throughput, trial-and-error screening or structure-based design, AI-driven models can generate and refine more efficacious molecular candidates with unprecedented speed and precision. By analyzing vast datasets, GenAI predicts binding interactions, optimizes key pharmacokinetic properties, and helps navigate the complex landscape of drug safety. This shift is enabling entirely new strategies for tackling diseases without suitable treatment options.
Key applications of GenAI in drug discovery include:
- Molecule optimization:5 This can be used to improve target binding and/or specificity and solubility, stability, aggregation, and other developability characteristics.
- De novo molecule generation:6 Unlike conventional drug design, which often starts with known chemical scaffolds, GenAI can create entirely novel structures tailored to specific biological targets. Diffusion models, transformers, flow-based models, and energy-based models can be used to explore vast chemical spaces, proposing new compounds with optimized binding affinity, solubility, and stability.
- Drug repurposing:7 AI can identify existing molecules with previously unrecognized therapeutic potential. By cross-referencing known drugs with disease-related molecular signatures, AI predicts alternative indications, offering a cost-effective route to new treatments.
- Biomarker discovery:8 AI-driven pattern recognition analyses complex data to uncover signatures associated with disease progression or treatment response. This is critical for advancing precision medicine and improving patient stratification in clinical trials.
- Targeting challenging proteins:9 AI is opening new therapeutic options for challenging proteins like GPCRs and ion channels, which are vital drug targets involved in multiple diseases but notoriously difficult to modulate. By designing epitope-specific antibodies that precisely bind to key regions, AI offers potential breakthroughs for diseases where conventional approaches have fallen short – increasing the likelihood of generating high-affinity candidates with improved clinical success rates.
Beyond molecule design, GenAI can improve efficiency and cost-effectiveness by filtering out weak candidates early in development. Instead of advancing less effective molecules through costly laboratory testing, AI models can predict potential failures before they reach preclinical stages. This data-driven triage reduces attrition rates and shortens drug discovery timelines.10
Real-world progress
While there have not been any de novo AI-designed drugs that have cleared Phase III trials and approved for use so far, AI is becoming increasingly adopted within drug discovery and is shaping industry practices. AI tools are accelerating compound selection, optimizing drug properties, and repurposing existing therapies, bringing a shift from theoretical ideas to real-world applications.
Companies already succeeding in using AI in drug discovery (see Table 1) include Healx, which recently dosed the first patient in a Phase 2 trial with HLX-1502, an AI-discovered therapy for rare tumour disorder neurofibromatosis type 1. And BenevolentAI used its proprietary AI platform to identify baricitinib, originally approved for rheumatoid arthritis, as a potential treatment for COVID-19.
Table 1: A selection of active AI drug discovery programs.
AI is also being applied to traditionally hard-to-drug targets, particularly through pharmaceutical partnerships with smaller, specialized AI-driven biotech firms to help accelerate drug discovery. In the past six months alone, there have been several pharma-biotech partnerships, including Nxera Pharma, who has partnered with my company, Cardiff-based techbio company Antiverse, to design novel GPCR-targeted antibody therapeutics using GenAI. Another example is Insilico Medicine, which has partnered with Tenacia Biotechnology in a project focusing on using generative AI for the early discovery stage of novel CNS disease therapies, developing small-molecule inhibitors from scratch.
If these AI-led methods continue to prove effective, the impact could be far-reaching. Generative models may unlock new treatments for major diseases such as cancer, heart disease, and neurodegeneration, where conventional therapies fall short.
Balancing hope with reality in AI drug discovery
While GenAI has the potential to reshape drug discovery, its success depends on the quality and availability of data. AI models rely on vast, diverse datasets to generate meaningful predictions, but data constraints pose a major challenge. Many biotech datasets are locked in data silos, forcing researchers to duplicate existing work instead of building upon it. This inefficiency slows progress and hinders the potential impact of AI.
Even when data is publicly available, quality and reliability are not guaranteed. Biases, errors, and incomplete reporting are common, particularly in cases where positive results are prioritized over replication studies. In drug discovery, where AI models must distinguish subtle molecular interactions, these inaccuracies can lead to misleading predictions.11
Without high-quality, biologically representative datasets, GenAI’s predictive power is limited, increasing the risk of false positives and ineffective drug candidates.12 As a result, rigorous in vitro and in vivo validation remains essential to confirm that AI-designed therapies not only work as predicted, but also produce meaningful therapeutic effects and lack harmful side effects.
Expanding what is scientifically possible
GenAI offers a realistic path forward in tackling previously undruggable targets, with the potential to reshape the entire therapeutic landscape and the whole drug discovery pipeline. By designing novel molecules, optimizing key attributes, and identifying strong candidates earlier, GenAI is accelerating drug discovery and also expanding what is scientifically possible.
However, the success of AI-driven drug discovery depends on more than just technological advancements. Data integrity must come first – before AI-generated molecules can advance through experimental validation and demonstrate real-world efficacy. Achieving this will require close collaboration between data scientists, medicinal chemists, biologists, and regulatory experts, ensuring that AI-driven insights translate into clinically meaningful breakthroughs, offering renewed hope to patients still waiting for effective treatments.
References
- Sun D, Gao W, Hu H, Zhou S. Why 90% of clinical drug development fails and how to improve it?. Acta Pharm Sin B. 2022;12(7):3049-3062. doi:10.1016/j.apsb.2022.02.002
- Spradlin JN, Zhang E, Nomura DK. Reimagining druggability using chemoproteomic platforms. Accounts of Chemical Research 2021;54(7):1801–1813. https://doi.org/10.1021/acs.accounts.1c00065
- US Congressional Budget Office. Research and Development in the Pharmaceutical Industry (2021). https://www.cbo.gov/publication/57126
- Zhavoronkov A. Caution with AI-generated content in biomedicine. Nature Medicine 2023;29:532.
- Chen Z, Wang X, Chen X, et al. Accelerating therapeutic protein design with computational approaches toward the clinical stage. Computational and Structural Biotechnology Journal. 2023; (21):2909-2926. doi: 10.1016/j.csbj.2023.04.027.
- Skwark MJ, Panić J, Elofsson A. (2023). Application of Generative AI in Drug Discovery. SpringerLink.
- Cortial L, Montero V, Tourlet S, et al. Artificial intelligence in drug repurposing for rare diseases: a mini-review. Front Med 2024;11:1404338.. doi:10.3389/fmed.2024.1404338
- Dana J, Venkatasamy A, Saviano A, et al. Conventional and artificial intelligence-based imaging for biomarker discovery in chronic liver disease. Hepatol Int. 2022;16(3):509-522. doi:10.1007/s12072-022-10303-0
- Chen Z, Ren X, Zhou Y, Huang, N. Exploring structure-based drug discovery of GPCRs beyond the orthosteric binding site. hLife 2024:2(5);211-226.
- Blanco-González A, Cabezón A, Seco-González A, et al. The Role of AI in Drug Discovery: Challenges, Opportunities, and Strategies. Pharmaceuticals (Basel). 2023;16(6):891. Published 2023 Jun 18. doi:10.3390/ph16060891
- Kozlov M. ‘Publish or Perish’ is now a card game — not just an academic’s life. Nature. 2024;632(483). doi:10.1038/d41586-024-02511-5
- Stanford Institute for Human-Centered AI. (2024). Data-centric AI: AI models are only as good as their data pipeline.
Murat Tunaboylu ([email protected]) is the co-founder and CEO of Antiverse, a start-up working on computational antibody design technology based in Cardiff, Wales.
The post How AI is Cracking Medicine’s Most Challenging Drug Targets appeared first on GEN - Genetic Engineering and Biotechnology News.