Drug discovery has always been a complex, costly, and time-intensive process, often stretching over a decade with high rates of failure. Now, advances in artificial intelligence (AI), machine learning (ML), and automation are beginning to reshape the earliest stages of discovery, offering new ways to accelerate timelines, reduce costs, and uncover novel therapeutics.

Central to this shift is the growing recognition that data quality and accessibility are just as important as the algorithms themselves. Without reliable, standardized, and connected data, even the most sophisticated AI models cannot deliver their full potential.

To better understand how these challenges and opportunities are playing out in practice, DDN spoke with Raymond Wong, Technical and Sales Manager at Shimadzu UK. Wong shared his perspective on how AI is changing the role of the laboratory bench, why breaking down data silos could unlock a new era of discovery, and how more open, collaborative approaches are the key to turning AI’s promise into real medicines.

How is AI changing the early stages of discovery?

Drug discovery often spans over a decade, with an alarming failure rate of around 90 percent for drug candidates in preclinical or clinical trials. This challenge is further exacerbated by the vast complexity and scale of scientific data involved, which has historically presented significant obstacles. Even with the aid of computational tools, the magnitude of the task has often outstripped the capabilities of available technologies.

However, we are now at a pivotal moment, driven by the emergence of transformative technologies like AI and ML. AI, in particular, can analyze vast and complex datasets at a speed and scale that would be impossible for humans. For instance, AI can identify patterns, predict molecular interactions, and prioritize the most promising compounds from millions of options. This not only accelerates the research process but also reduces costs and significantly enhances the probability of discovering effective treatments.

One of the most revolutionary aspects of AI in drug discovery is its capacity to create a continuous, iterative improvement loop. Here’s how it works: Data from laboratories and clinical trials are used to train AI models and algorithms. These models then generate predictions about drug targets, therapeutic molecules, and other key areas. The predictions are subsequently tested in the lab, producing new data that feeds back into the system to refine and improve the models further. Over time, this iterative process enhances accuracy and efficiency, bypassing the traditional trial-and-error bottlenecks that have long hindered progress in developing new therapies.

Even more exciting is AI’s potential to automate the synthesis of drug targets and the generation of data, creating a self-improving ecosystem. This approach could allow AI to uncover entirely new therapeutics that might not have been conceived through traditional methods. This isn’t just incremental progress, it represents a paradigm shift in how we approach drug discovery and development.

How do you see the balance between computational and experimental approaches evolving?

Traditionally, wet chemistry has been the backbone of experimental research, with hundreds of chemists conducting thousands of reactions daily. However, advancements in computational simulations mean that AI can process billions of virtual reactions in mere minutes.

Rather than being the primary driver of experimentation, wet chemistry may increasingly serve as a tool for confirmation.
- Raymond Wong

This capability is further enhanced by integrating predictive AI models into Design-Build-Test-Learn cycles, where billions of virtual compounds are screened, optimized, and iteratively refined. Autonomous laboratories, which are becoming more prevalent, play a key role in this process by identifying and optimizing promising candidates to feed back into AI/ML models, creating a continuous loop of improvement.

Looking ahead, I see wet chemistry shifting its focus. Rather than being the primary driver of experimentation, it may increasingly serve as a tool for confirmation, validating the predictions and insights generated by computational methods.

What data challenges do customers face when adopting AI approaches?

One key bottleneck our customers face in adopting AI/ML approaches is the challenge of generating high-quality, comprehensive datasets to feed these models. While public data from publications is often filtered and typically excludes failed datasets, successful AI/ML models rely on all types of data, including failures, to refine and optimize predictions.

Shimadzu addresses this by supporting early adoption of automated platforms that generate real, unfiltered data. Tools like LabSolutions Sync equip laboratories with the ability to efficiently schedule and manage chromatographic and mass spectrometry systems, producing essential datasets in clean, unified formats such as Allotrope™. Additionally, integrating automated sample preparation systems, like liquid handlers or cooperative robots, ensures efficient feeding of samples into analytical systems.

As automation evolves, laboratories must also consider scheduling systems with sufficient logic to handle day-to-day challenges, such as optimizing sample distribution to online systems. By focusing on these aspects, organizations can overcome data bottlenecks and fully leverage AI/ML in their discovery pipelines.

Industry silos and incompatible data formats are often cited as key obstacles to realizing the full potential of AI/ML in drug discovery. From your perspective, what needs to change?

The main challenge lies in the way data is traditionally stored and processed. Many organizations still operate in silos, where data is analyzed in a linear, unimodal fashion, one modality at a time. This approach is inherently limited because it doesn’t allow for the integration of diverse data types, such as cell data, molecular data, clinical records, imaging data, ADME (Absorption, Distribution, Metabolism, and Excretion)-Tox profiles, and even unstructured text like publications or clinical trial protocols.

To address this, we need to shift towards a multimodal approach to data architecture. Multimodality enables the integration and simultaneous analysis of multiple data sources, creating a holistic view of the problem. For example, combining genomic, chemical, clinical, and imaging data allows us to detect and connect trends that would otherwise remain hidden in isolated datasets. This not only enhances interpretability but also builds trust among regulators, researchers, and industry stakeholders by providing more comprehensive and reproducible insights.

However, achieving this requires significant changes. Firstly, data formats need to become standardized and interoperable across organizations and platforms. This means moving away from proprietary and inflexible systems to ones that are designed for seamless data sharing and integration. Secondly, the industry needs to invest in creating unified, high-quality knowledge bases that address the inherent heterogeneity and inconsistencies in biomedical data. Only then can we effectively fuel large language models and other AI tools with the kind of robust, multimodal data they need.

Ultimately, the goal is to create a data ecosystem that is not just clean and connected but also dynamic, one that evolves as new data is generated. By breaking down silos and embracing multimodality, we can unlock the full potential of AI/ML to accelerate drug discovery and deliver better outcomes for patients.

Shimadzu has made strides toward more open and collaborative solutions, such as enabling external AI/ML models to access data from its chromatography systems. Could you share a bit more about this approach and why it matters?

Shimadzu has taken significant steps to foster more open and collaborative solutions in drug discovery, particularly through enabling external AI/ML models to access data from its chromatography systems. Chromatographic analytical data is crucial in pharmaceutical research, providing reliable insights into a compound’s structure, purity, potency, and behavior, essential for every stage of development from early research to manufacturing.

Traditionally, proprietary software has restricted instrument control and data accessibility, limiting the integration of AI/ML capabilities. Shimadzu’s LabSolutions Sync addresses this by allowing third-party direct instrument control and supporting output in the Allotrope data format, which standardizes data across manufacturers and software platforms. This ensures seamless data processing and compatibility. Additionally, raw data and PDF reports from various instruments can be automatically imported into the LabSolutions database, enabling secure and efficient data management.

By removing these barriers, Shimadzu is enabling laboratories to adopt AI/ML models more effectively, facilitating both instrument control and data generation to refine predictive models and accelerate the discovery of new therapeutics.

Are there any real-world examples or case studies you can share where Shimadzu’s tools or systems have played a role in advancing AI-powered drug discovery?

A great example of Shimadzu’s role in advancing AI-powered drug discovery is its collaboration with Exscientia, now part of Recursion. Exscientia made headlines as the first company to bring an AI-designed drug candidate into clinical trials in 2020, and has since accelerated drug development timelines, delivering candidates in just 12 to 15 months — far faster than traditional methods.

While Exscientia is renowned for its AI innovations, it has also pioneered laboratory automation to streamline the synthesis, purification, and testing of drug candidates. Shimadzu’s analytical tools have been integral to this effort. For instance, Shimadzu’s ultra-fast preparative Liquid Chromatography (LC) with fraction-trapping capabilities eliminated the lengthy dry-down process in isolating target compounds, a critical advancement for Exscientia’s workflows.

Additionally, Shimadzu’s ability to offer both LC and Supercritical Fluid Chromatography (SFC) techniques simplified the creation of Exscientia’s integrated automation platform. The reliability of Shimadzu’s equipment and its LabSolutions Sync software provided the flexibility needed to seamlessly integrate Exscientia’s in-house informatics systems.

What truly sets Shimadzu apart, however, is its collaborative mindset. Exscientia required bespoke configurations and software to build a novel, integrated setup, and Shimadzu’s willingness to innovate quickly and work closely with the team was key to the project’s success. This partnership exemplifies how Shimadzu’s tools and collaborative approach are driving forward AI-powered drug discovery.

The integration of Exscientia into Recursion creates a highly complementary partnership, combining expertise to deliver an end-to-end drug discovery approach that spans from biology through to chemistry.

This interview has been condensed and edited for clarity.