Cluster Synthesis

How is artificial intelligence being utilised in labs to synthesise design and workloads?

Yann Gaston-Mathé at Iktos

While artificial intelligence (AI) now accelerates molecular design at unprecedented speed, laboratory synthesis continues to lag behind, limiting the impact of digital innovation. Emerging frameworks such as AI-driven orchestration and cluster synthesis offer a path to reshape how laboratories align design with execution.

In a period marked by rapid advances in AI, drug discovery is undergoing profound change. Models such as AlphaFold 3, capable of predicting full biomolecular complexes with striking accuracy, and the emergence of compute infrastructures like BioHive-2 – which signal the industrialisation of model training in chemistry and biology – are unlocking design capabilities previously out of reach.1,2 Yet, amid this acceleration, one constraint remains strikingly constant: the ability to synthesise compounds efficiently and at scale.

In drug discovery, no matter how powerful our digital models become, real progress ultimately depends on molecules that can be made and tested. Physical synthesis is the indispensable bridge between AI-generated hypotheses and the experimental data required to refine them. When synthesis is slow, the entire discovery process slows with it, delaying time to candidate selection. As highlighted by a recent opinion letter from the Roche Innovation Center in Basel, Switzerland, synthesis remains the primary bottleneck within the Design-Make-Test-Analyse (DMTA) cycle, particularly for advanced and complex molecules.3

Recent developments illustrate this industry-wide tension. Eli Lilly’s decision to sell one of the world’s most advanced automated synthesis labs – followed a year later by a major investment in AI supercomputing through a partnership with NVIDIA – sparks reflection on a strategic challenge discovery leaders must confront: AI without an equally agile synthesis capability cannot deliver its full value.4,5 Automated synthesis platforms have struggled to keep pace when required to produce the chemically diverse series needed for effective exploration of chemical space, the development of innovative first-in-class therapeutics and the establishment of strong intellectual property positions. Increasing internal synthesis capacity – or outsourcing to clinical research organisations (CROs) – does not provide a scalable solution either as it fails to create the reproducible acceleration required to keep pace in such a competitive landscape. For forward-looking organisations, the question is no longer whether to adopt AI, but how to ensure that the laboratory can keep pace. An emerging answer is AI-driven orchestration – integrating de novo design, synthesis planning, automation technologies and scheduling intelligence into a unified, responsive workflow. By coordinating these once-siloed steps, discovery teams can move more programmes in parallel, shorten DMTA cycles, and strengthen the competitive advantage that comes from rapid iteration and data-driven decision-making.

The two sides of the modern efficiency gap

Generative models can now propose thousands of high-quality molecular structures in hours. Yet, many of these ideas sit in regions of chemical space that are difficult – or sometimes effectively impossible – to synthesise. Even with the growing use of synthetic-accessibility predictors for compound prioritisation, proposed compounds may still require lengthy multi-step routes, creating downstream bottlenecks that undermine the value of rapid design cycles. This disconnect has organisational consequences: while computational chemists readily embrace AI-enabled exploration, medicinal and synthetic chemistry teams often face feasibility barriers that weaken confidence in the outputs, leading to disparate technology adoption across departments.

On the execution side, automation has not kept pace with this surge in design diversity. For decades, chemistry was driven primarily by manual synthesis – highly skilled, but inherently sequential and limited in capacity. Automation was introduced to expand throughput, enabling high-throughput optimisation of a single transformation or systematic library synthesis around a shared scaffold; approaches that excel when workflows are repetitive and uniform. As synthesis demands grew, organisations increasingly turned to CRO networks – both specialised onshore groups and large offshore operations – to further expand throughput and absorb routine or labour-intensive chemistry.

Yet, neither outsourcing nor robotics addresses the core challenge. Robotic platforms remain constrained by rigid, transformation-specific workflows, and CRO reliance expands capacity but introduces variability, long cycle times and other outsourcing risks. As a result, the design-make gap widens; ideas accelerate, but synthesis remains anchored in processes optimised for uniformity, not diversity. This imbalance limits how quickly experimental data can be generated, fed back into models and used to guide the next round of design – reinforcing synthesis as the rate-limiting step in the DMTA cycle.

The new framework: from automation to AI-driven orchestration

Recognising these limitations, the industry can shift from isolated processes across Design, Make and Test to a fully orchestrated model, where de novo design, retrosynthesis planning, laboratory automation and scheduling intelligence operate as one coordinated system. AI-driven orchestration replaces linear hand-offs with a dynamic workflow in which each step is shaped by real-world constraints such as target product profile requirements, synthetic feasibility, material availability, equipment capacity and operational timing. In this orchestrated framework, molecular design is guided not only by traditional criteria such as potency, selectivity or novelty, but by the practical requirements of robotic synthesis – ensuring that ideas entering the queue are compatible with what an automated laboratory can execute. Retrosynthesis planning moves beyond generating theoretical routes to prioritising pathways that balance chemical tractability with campaign-level efficiency, enabling automated systems to execute routes consistently and at scale.

The broken link: updating automation in the era of AI

To realise the orchestration vision, laboratory automation must evolve beyond reaction-class homogeneity.

High-throughput experimentation and scaffold-based library synthesis excel when reactions share the same transformation, solvent, heating profile or work-up. These approaches deliver efficiency through repetition, governed by rigid operation systems – fixed temperature profiles, static workflows and predefined resources – meaning they function well for incremental analogue expansion but struggle when projects require fundamentally different chemistries in the same cycle.

These limitations are amplified by the practical realities of laboratory operations. Automated platforms must deal with inventory availability, limited reactor slots, powder-dispensing rules, solvent compatibility and procurement delays; yet most scheduling systems do not integrate these constraints. As a result, robotic capacity often remains under-utilised or locked to a single project at a time, even when multiple teams could benefit. While AI has transformed reaction optimisation and predictive retrosynthesis, far fewer developments have focused on how to orchestrate many different reactions across diverse projects in a single automated run.

Cluster synthesis: unlocking chemical diversity on robotic platforms

Latest developments in the field now open the door to a paradigm shift in high-throughput robotic synthesis, moving from mono-reaction-type libraries to multi-reaction-type clusters.6 Rather than grouping reactions solely because they share the same scaffold or transformation, cluster synthesis groups them according to compatible operating windows, most often in temperature and time. When those ranges overlap, reactions from entirely different families can be executed in parallel on the same automated platform. New AI-based scheduling and planning models emerge to help chemists organise the workload into the minimum number of feasible clusters, integrating both chemical compatibility and the physical constraints of the robotic system.

Cluster synthesis reframes automated synthesis from a tool that supports one project at a time into adaptable infrastructure capable of advancing multiple programmes and chemotypes simultaneously, offering a scalable pathway to close the gap between accelerated design and synthesis.

What cluster synthesis enables in practice

When applied systematically, cluster synthesis unlocks capabilities that traditional automated chemistry cannot achieve:

Higher experimental diversity per campaign
The increased diversity dramatically accelerates synthetic-aperture radar generation, enabling rapid prioritisation of chemical series and providing richer data to guide design. A single automated run can span many chemotypes and transformation types, allowing teams to test fundamentally different hypotheses at once and improving the likelihood of identifying viable hits for challenging targets.

Greater responsiveness to shifting project needs
Clusters are built around condition compatibility and real-world feasibility – available building blocks, solvent constraints, reactor capacity and procurement timelines. This ensures synthesis continues even when materials are delayed or priorities change, keeping DMTA cycles moving and maintaining project momentum.

Scalable campaigns aligned with discovery demand
Because clusters grow by incorporating reactions that share compatible ranges – not identical chemistries – campaigns can expand naturally as design output increases.

This provides a scalable way to manage rising design volume without requiring bespoke optimisation for every new scaffold.

Cross-project execution on shared robotic infrastructure
Cluster synthesis enables reactions from different programmes to run in the same campaign, reducing queues and increasing platform utilisation. Robotic systems shift from single-project workhorses to shared assets that support broader portfolio progression.

Human-guided, machine-executed workflows
Chemists define reaction templates – set of conditions, optional steps and operational rules – while algorithms assemble clusters, resolve conflicts and optimise execution. This preserves scientific judgment while enabling automation to handle complexity at scale, resulting in rigorous, traceable and operationally efficient synthesis cycles.

More predictable and data-rich synthesis
Structured clusters and compatibility rules produce more interpretable outcomes across diverse reactions. As data accumulates, patterns become clearer, informing route selection, improving predictability, and strengthening future design and synthesis decisions.

An important part of this new synthesis paradigm is the role of scheduling intelligence, which acts as the critical interface between design outputs and robotic execution. As diverse compounds enter the make pipeline, AI-based scheduler evaluates feasibility, resource constraints and condition compatibility, proposing the most efficient clustering strategy for each campaign. Yet, human oversight remains central; chemists review these recommendations, adjust priorities and ensure alignment with project goals. This collaboration allows experts to guide which hypotheses move forward while delegating logistical complexity to the scheduler. In this way, AI enables teams to manage increasingly complex and diverse synthesis workloads with clarity, control and speed.

Conclusion and outlook: turning design power into discovery speed

AI has dramatically expanded what can be imagined and designed in silico, but the impact of these advances will only be realised when laboratories can make molecules with equal agility. For discovery organisations striving to accelerate timelines, improve decision quality and enhance portfolio productivity, two shifts stand out: AI-driven orchestration and cluster synthesis.

Together, these approaches align design with execution, bringing synthesis closer to the speed and flexibility required by modern AI-driven R&D. Orchestration provides the connective tissue linking Design, Make and Test into a unified process; cluster synthesis offers the operational flexibility that makes this integration feasible on real laboratory infrastructure. Organisations that embrace this combined framework will be positioned to learn faster, iterate more effectively and push more programmes towards candidate selection with greater predictability.

References:

1. Jumper J et al (2024), ‘Accurate structure prediction of biomolecular complexes with AlphaFold 3’, Nature, 630, 493-500
2. Visit: blogs.nvidia.com/blog/drug-discovery-recursionsupercomputer/
3. Visit: pubs.rsc.org/en/Content/ArticleLanding/2025/MD/ D5MD00672D
4. Visit: fiercebiotech.com/cro/eli-lilly-sells-science-studiolab-arctoris-moving-automation-platform-san-diegooxford
5. Visit: blogs.nvidia.com/blog/lilly-ai-factory-nvidiablackwell-dgx-superpod/
6. Visit: chemrxiv.org/engage/chemrxiv/article-details/68d e33daf2aff167708137a8