In a paper published in the journal Nature Communications, researchers presented a groundbreaking framework for distributed self-driving laboratories within the world avatar project. This initiative aimed to create a comprehensive digital twin using a dynamic knowledge graph, facilitating global collaboration for scientific advancements.
Study: A Framework for Distributed Self-Driving Laboratories in the World Avatar Project. Image credit: Gorodenkoff/Shutterstock
Utilizing ontologies and autonomous agents, the framework captured and executed experimentation workflows while ensuring meticulous data provenance. The practical application involved connecting robots in Cambridge and Singapore for collaborative closed-loop optimization, exemplified by a pharmaceutically relevant aldol condensation reaction. The autonomous evolution of the knowledge graph aligned with the researchers' goals, demonstrating its efficiency in generating a Pareto front for cost-yield optimization within three days.
Related Work
The concept of self-driving laboratories (SDLs), evolving from the historical roots of laboratory automation since the 1960s, has significantly impacted scientific domains such as chemistry, materials science, biotechnology, and robotics. While SDLs have led to accelerated discoveries, their implementation has posed challenges, often requiring specialized teams and centralized setups.
Researchers recognize the imperative for a global collaborative research network, initiating efforts to decentralize SDLs and seamlessly integrate diverse research groups. This paradigm shift faces challenges in efficiently orchestrating heterogeneous resources, sharing data across organizations, and ensuring data provenance recording.
Previous attempts have addressed these challenges with middleware, data-sharing protocols, and provenance initiatives. Semantic web technologies, specifically knowledge graphs like the world avatar, emerge as a promising solution, providing a common language for immediate data dissemination between SDLs and addressing challenges faced in earlier studies.
SDL Development and Implementation
In adherence to the best practices of the world avatar project, this study meticulously follows the version control system on GitHub for all ontologies and agents. The researchers designed the approach to capture the thought process behind the development, with principles that individuals can extend to self-optimization applications in various domains.
The ontological development process, inspired by relevant software tools and consultation with domain experts, involves an iterative journey from specifying target deliverables to conceptualizing pertinent concepts and implementing code for queries. Researchers crafted the ontologies of the world avatar to be comprehensible by software agents, drawing inspiration from existing reaction database schemas and relevant software tools.
Following ontological development, researchers define agents as executable entities that process inputs and generate outputs, with their input/output signatures represented following the ontogent. Implementing these agents is facilitated by the derivation agent template in Python, provided by the derived information framework. This framework handles asynchronous communication with the knowledge graph, where agents monitor assigned tasks and record execution progress. Researchers conduct unit and integration tests to ensure responsible development, employing simulations to verify data flows in distributed SDLs.
Researchers draw inspiration from remote control practices in lab automation to design the knowledge graph, ensuring it spans the internet. They follow deployment practices standard to cloud-native applications and implement them through docker containers. The triplestore and file server, housing knowledge statements, are deployed at internet-resolvable locations, while agents, based on capabilities, are distributed across different host machines.
Security considerations mandate deploying agents that monitor and control hardware in their respective laboratories. Agents autonomously register their ontoagent instances in the knowledge graph at startup, forming a distributed network facilitating information transfer within the knowledge graph and bridging the gap between cyberspace and the physical world.
Connecting two automated flow chemistry platforms in Cambridge and Singapore, the study employs similar methodologies with slight variations in experimental setups. Both setups involve vapourtec R2 pump modules and a vapourtec R4 reactor module, yet the liquid handling mechanisms differ. The Cambridge setup employs a Gilson liquid handler, while Singapore uses reagent bottles directly attached to vapourtec pumps. Detailed descriptions of the experimental setups, including chemical compositions and high-performance liquid chromatography (HPLC) analyses, highlight the uniformity and distinct characteristics of the two interconnected SDLs.
Revolutionizing Distributed SDLs Architecture
The study introduces a conceptual architecture for distributed SDLs that revolutionizes closed-loop optimization through design-make-test-analyze (DMTA) cycles. By seamlessly orchestrating both computational and physical resources, SDLs facilitate the integration of data and material flows, effectively bridging the virtual and physical realms.
This proposed architecture empowers scientists to establish research goals, initiating a closed-loop process in cyberspace, wherein monitoring components interpret goals, iterations accumulate prior information, and a design component suggests physical experiments. This system autonomously traverses cyber and physical spaces within a dynamic knowledge graph, efficiently achieving research objectives or managing resources within allocated limits.
While the study's architecture liberates scientists from routine tasks, challenges emerge concerning implementation robustness, scalability, maintainability, safety, and ethics. The adoption of dynamic knowledge graph technology is proposed as a solution, abstracting software components into agents and enabling real-time control in cyberspace. The three layers of the knowledge graph, representing the real world, a dynamic knowledge graph in cyberspace, and active agents facilitating real-world changes, redefine closed-loop optimization as information dynamically traversing the knowledge graph. This transformative approach establishes a robust framework for authentic distributed SDLs.
Conclusion
In conclusion, the study introduces a groundbreaking conceptual architecture for distributed SDLs, redefining closed-loop optimization through DMTA cycles. Researchers designed a system operating within a dynamic knowledge graph that autonomously navigates both cyber and physical spaces, effectively accomplishing research objectives while addressing implementation challenges.
The study proposes adopting dynamic knowledge graph technology, envisioning a transformative framework for authentic distributed SDLs that provides flexibility and extensibility and enables real-time data sharing across organizations. It marks a significant stride towards the future of autonomous and integrated laboratory workflows, paving the way for continuous innovation in lab automation.
Can We Ethically Navigate AI’s Future and Risks?