A laboratory designed around AI
A research group at Shanghai Jiao Tong University has built a scientific laboratory described as “AI-native,” named AutoDNA (or Siyuan-1), capable of conducting molecular biology experiments without direct human intervention. The news is reported by IT之家, which cites the publication of the study in the journal Nature, where the work was featured at the top of the website. The research involved the IPADS laboratory of the School of Computer Science, with the teams of Chen Haibo and Wang Zhaoguo, and the groups of Fan Chunhai and Wang Fei of the School of Chemistry and Chemical Engineering and the Center for Transformative Molecular Sciences, with support from the DNA storage research center of the Zhangjiang Institute for Advanced Study.
According to the source, the novelty is not the use of AI in individual stages, but the joint design of model, experiment and instrumentation. The stated goal is to overcome the limitations of existing autonomous facilities when dealing with molecular biology: organization of complex processes, coordination among different instruments, and ensuring timing, environmental stability and operational precision. The claimed result is coverage of scenarios requiring micrometric positioning and microliter-level liquid handling.
From “add-on” to integrated system
IT之家 explains that existing autonomous laboratories generally operate with a “plug-in” architecture: experts configure the instruments in advance, define the procedures and establish optimization rules, while AI intervenes only in segments such as literature search, protocol generation or data analysis. This approach can make a single step more efficient, but it struggles to handle the complexity of biomolecular procedures and the variety of equipment.
AutoDNA instead adopts an AI-native architecture. The system breaks the experimental task into multiple stages — literature search, reagent selection, protocol design, code generation, hardware execution, results analysis and iterative optimization — and entrusts multiple specialized agents with coordinating them based on the progress of the experiment. In this way, according to the source, the model can follow the entire “design-execution-optimization” cycle and modify subsequent steps based on the results obtained.
As for the instruments, the capabilities of each device are translated into “atomic services” described in natural language and with uniform code interfaces, so that the model knows what each machine can do and how to activate it. This abstraction layer hides differences between models and makes it possible to generate, verify and correct multi-device control code. On the precision front, the automated hardware is accompanied by an execution and planning system that controls programs, tracks instrument status and allocates resources, so that multiple activities proceed in compliance with time and operational constraints.
Applications and open questions
The platform also integrates a control terminal based on the OpenHarmony operating system. IT之家 reports that application tests have been presented in areas such as nucleic acid testing, enzymatic DNA synthesis and cancer diagnosis; the published figure shows AutoDNA autonomously completing the entire workflow of encoding, writing, storage, reading and decoding for DNA storage.
However, some aspects not covered by the available material remain to be clarified: no quantitative experimental results, success rate compared with human operators, construction costs or development times are indicated. The news is based on what the researchers stated and the source reported, and independent verification of performance will require reading the full text of the study in Nature.
