Hierarchically Encapsulated Representation for Protocol Design in Self-Driving Labs

TL;DR

This work tackles the challenge of rapid, reliable protocol design in self-driving laboratories by introducing a hierarchically encapsulated representation implemented as a domain-specific language (DSL) that unifies three abstraction levels: instance actions with attributes, operation-centric function abstraction, and product-flow-centric model abstraction. It proposes automatic representation generation via hierarchical non-parametric modeling to tailor the DSL to domain corpora and enable domain-specific protocol design. The authors demonstrate that dual-view representations with external verification (Encapsulated+ with External) outperform baselines across genetics, medicine, bioengineering, and ecology, suggesting a viable path for complementing LLMs in machine-assisted scientific exploration. The framework has implications for scalable, cross-domain protocol design and points to future directions such as digital twins and more general cross-domain applicability in self-driving laboratories.

Abstract

Self-driving laboratories have begun to replace human experimenters in performing single experimental skills or predetermined experimental protocols. However, as the pace of idea iteration in scientific research has been intensified by Artificial Intelligence, the demand for rapid design of new protocols for new discoveries become evident. Efforts to automate protocol design have been initiated, but the capabilities of knowledge-based machine designers, such as Large Language Models, have not been fully elicited, probably for the absence of a systematic representation of experimental knowledge, as opposed to isolated, flatten pieces of information. To tackle this issue, we propose a multi-faceted, multi-scale representation, where instance actions, generalized operations, and product flow models are hierarchically encapsulated using Domain-Specific Languages. We further develop a data-driven algorithm based on non-parametric modeling that autonomously customizes these representations for specific domains. The proposed representation is equipped with various machine designers to manage protocol design tasks, including planning, modification, and adjustment. The results demonstrate that the proposed method could effectively complement Large Language Models in the protocol design process, serving as an auxiliary module in the realm of machine-assisted scientific exploration.

Figures (5)

• Figure 1: The representations for protocol design.(A) The example of protocol design by novice and veteran experimental scientists. (B) The hierarchies of our proposed representation, from original full protocol representation, to dual representation of operation- and product-flow-centric views.
• Figure 2: Diagram of automatic representation generation.(A) Illustration of the workflow. (B) Convergence curve of automatic function abstraction. (C) Convergence curve of automatic model abstraction. (D-F) Confusion matrices on operation distribution (D), product distribution (E), and device distribution (F), between dsl across domains. Correlation scores are low except the ones along the diagonals, indicating the significant inter-domain distinctions between the resulting dsl.
• Figure 3: Results of protocol design.(A) Profile of text-level similarity between testing sets of the three tasks. (B) Pairwise comparison between the capabilities of different machine designers across the six dimensions. (C-E) Performances of the seven machine designers on the planning (C), modification (D), and adjustment (E) tasks across the six dimensions (index by column).
• Figure A1: Comparison between the perplexity of the test set and the reference set
• Figure A2: Visualization of diversity between novel protocols