AI drug discovery in pharma: Partnerships and pitfalls

Key Takeaways

• Accelerated Discovery: AI is streamlining drug development through automated target identification, molecular simulations, and toxicity prediction.
• Industry Partnerships: Major players like AstraZeneca, Pfizer, and Janssen are collaborating with AI specialists to optimize clinical trials and molecular modeling.
• Proven Milestones: AI-designed molecules have already entered human clinical trials, and tools like AlphaFold have mapped nearly all human proteins.
• Critical Challenges: Data quality, algorithmic bias, and the need for "Explainable AI" (XAI) remain primary hurdles for widespread adoption.
• Human-AI Synergy: AI is positioned as a complementary tool to enhance, rather than replace, traditional experimental methods and human expertise.

AI applications in drug discovery

Over the past several years, AI has moved from experimental to operational within drug development. Several foundational milestones have defined this trajectory:

• Exscientia celebrated the first-ever AI-designed drug molecule entering human clinical trials.
• DeepMind's AlphaFold predicted protein structures for 330,000 proteins, including all 20,000 proteins in the human genome, representing a substantial advance in structural biology.
• Insilico Medicine initiated Phase I clinical trials for the first-ever AI-discovered molecule based on an AI-discovered novel target.
• AbSci became the first entity to create and validate new antibodies using generative AI.
• The FDA granted its first Orphan Drug Designation to a drug discovered and designed using AI.

Since these early achievements, the field has accelerated further. Large language models and generative AI are now being applied across the full development workflow, from molecular generation to clinical trial design, shifting AI from a point solution to an integrated capability across the pipeline.

AI is currently being used across multiple different aspects of drug development, supporting the work of both biotechs and pharma, as well as other areas of the healthcare sector. AI is having a particular impact in the following areas:

Target identification: AI is trained on extensive datasets, including omics, phenotypic data, and disease associations, to understand biological mechanisms and identify new proteins or genes relevant to drug development. This matters because pharmaceutical research often clusters around the same well-known targets, limiting breakthrough potential. AI enables teams to systematically evaluate thousands of candidates in parallel, combining literature mining, human genetics data, and single-cell experiments to identify novel targets that traditional approaches would miss. Critically, the most effective applications combine computational screening with wet-lab validation, using AI to narrow the field before experimental confirmation.

Molecular simulations: Reducing the need for physical testing, AI enables high-fidelity molecular simulations entirely in silico, reducing dependency on physical screening and lowering early-stage development costs.

Prediction of drug properties: AI systems help predict key properties like toxicity, bioactivity, and physicochemical characteristics of molecules. There are numerous examples of this, such as:

• A machine learning (ML) algorithm called DeepTox outperformed all other methods by identifying specific features within molecules and could efficiently predict the toxicity of a molecule based on predefined features.
• The Similarity Ensemble Approach (SEA) was used to evaluate the safety target prediction of 656 marketed drugs against 73 unintended targets that might produce adverse effects.
• eToxPred was developed using an ML-based approach and applied to estimate the toxicity and synthesis feasibility of small organic molecules. It showed accuracy as high as 72%.

Drug design and repurposing: AI is able to find new drug candidates, either through screening existing molecules or designing entirely new ones. For example, a deep learning algorithm has been trained on a dataset of known drug compounds to propose new therapeutic molecules. AI also contributes to theoretical drug design by suggesting synthesis pathways for hypothetical drug compounds. What makes this particularly valuable is the multi-property optimization challenge. In traditional drug design, improving one property, such as potency, can compromise another, such as stability or metabolism, and teams must juggle 20 or 30 properties simultaneously. Generative AI approaches now enable rational molecule design based on protein structure, large-scale digital screening, and simultaneous optimization across multiple properties. Structure-based methods that leverage machine learning and three-dimensional molecular structure are also advancing, enabling more precise compound-target interactions. In some cases, AI can suggest modifications to compounds that make them easier to manufacture, further reducing development timelines.

Candidate prioritization: Once promising compounds are identified, AI systems can rank and prioritize them for further assessment based on predicted efficacy, safety profiles, and synthesizability, reducing the manual effort required in early-stage screening.

Predicting safety and side effects: Safety prediction is now recognized as one of the three critical steps in drug discovery being reshaped by AI. AI is being used to predict side effects for certain drugs and to determine optimal dosages for drug combinations, reducing the risk of adverse events. Beyond the discovery phase, AI is also being applied to post-market surveillance, enabling earlier detection of safety signals once drugs are in wider use. This end-to-end safety coverage, from molecular screening through real-world monitoring, represents a meaningful shift in how safety is managed across the development lifecycle.

Supporting clinical trials: AI contributes to clinical trials at multiple stages. In recruitment, it enables more precise identification of eligible patients through analysis of clinical, genetic, and real-world data. In trial design, it supports protocol optimization, site selection, and enrollment forecasting. During execution, wearable AI devices and remote monitoring tools provide real-time data streams that reduce site burden. AI is also used to identify patient subgroups most likely to respond to specific treatments, which is particularly relevant for genetically stratified and precision medicine studies where matching the right patient to the right therapy determines trial success.

AI collaborations in drug development

The most consequential advances in AI-driven drug development are emerging not from any single organization, but from structured partnerships between pharmaceutical companies, biotechs, and AI specialists. The following examples illustrate how these collaborations are translating into tangible pipeline progress.

Novartis: Novartis has applied AI-driven simulations to systematically evaluate thousands of gene candidates in autosomal dominant polycystic kidney disease (ADPKD), a condition where traditional approaches had struggled to identify viable targets. By combining large-scale computational screening with experimental validation in kidney organoids, the team narrowed the field to five promising targets in under a year. This example illustrates the emerging model: AI accelerates the search space, and wet-lab experimentation confirms the results, with neither step replacing the other.

Challenges in implementing AI in drug development

While AI is already contributing to drug development in meaningful ways, its integration introduces a set of challenges that are not trivial to resolve. One primary constraint is the availability and quality of data required to train AI models. AI-based approaches demand extensive, well-curated datasets, but in many therapeutic areas, the accessible data remains limited, incomplete, or inconsistently formatted. This directly affects the reliability of AI-driven outputs and the confidence that development teams can place in them.

Ethical considerations are also a major discussion point at the moment. Main concerns surround fairness and bias, particularly when training data is skewed or not representative or real-world populations, which it often is in clinical research. The potential implications of biassed AI programs include a range of negative effects, from inaccurate drug development decisions to unequal access to medical treatments. To address these ethical concerns, some are using strategies such as data augmentation (generating synthetic data to supplement existing datasets) and the use of explainable AI (XAI) methods to provide interpretable and transparent explanations for the predictions made by ML algorithms.

The collaborative structures being established between pharmaceutical companies, biotechs, and AI specialists are creating shared infrastructure and knowledge that will compound over time. However, the most effective programs treat AI as a complement to experimental methods rather than a replacement. In practice, this means AI narrows the search space, identifies patterns, and accelerates hypothesis generation, while wet-lab experimentation and clinical expertise validate and refine those outputs. As the Wellcome Trust has noted, recognizing AI's role as complementary is essential to realizing its potential. The operational challenge lies in building workflows that integrate computational and experimental steps seamlessly, so that AI-generated insights translate into reliable development decisions.

Conclusion

AI is reshaping drug development across the full lifecycle, from target identification and molecular design through clinical studies and post-market surveillance. The partnerships forming between pharmaceutical companies, biotechs, and AI specialists are accelerating this shift, producing tangible pipeline outcomes rather than theoretical potential.

The remaining challenge is not whether AI can contribute, but how to integrate it effectively into existing workflows. Data quality, ethical rigor, and the balance between computational and experimental methods will determine which organizations realize the full value of these tools.

For precision medicine specifically, AI is creating new opportunities to identify patient populations, refine eligibility criteria, and connect molecular insights to clinical execution. As these capabilities mature, the ability to link AI-driven discovery with structured patient engagement becomes increasingly important.

To learn more about how Sano supports precision medicine trials with integrated recruitment, genetic testing, and patient engagement, get in touch.