Student Johns Hopkins University Naperville, Illinois, United States
Introduction: Recent advancements in automation and machine learning have sparked a new method of conducting scientific research: self-driving laboratories. In a self-driving laboratory, robots, instead of laboratory technicians, execute experimental protocols and machine learning algorithms use previously collected data to design new experiments for the robots. This “closed loop” approach is especially useful for biological experiments in which the search space is large, meaning researchers must search through many combinations of reagents before finding the optimal outcome. In the realm of antibiotic drug discovery in developing countries, this problem becomes even more pronounced, as there are several both existing and novel treatments to be tested against the many antibiotic resistant bacterial strains. Additionally, implementing these large-scale protocols requires significant educational training and funding, which is more taxing on a developing country's resources. With a self-driving laboratory, science can be accelerated on two fronts: higher throughput experiments can be conducted using robotics and AI can be used to intelligently predict the next experimental iteration. Despite the promise of self-driving laboratories, few fully automated experiments have been performed in the biological space. This experiment seeks to execute a fully automated biological growth assay with automated data processing to determine whether a bacterial cell is susceptible or resistant to an antibiotic. The results from this paper will be from an automated growth assay for E. coli cells treated with tetracycline antibiotic. The results align which previous literature which supports E. coli susceptibility.
Materials and
Methods: There were 6 robots used in the automated growth assay workflow: Azenta Life Sciences Automated Plate Seal Remover, Hidex Sense Microplate Reader, Azenta Life Sciences Automated Roll Heat Sealer, Hudson EX Robotic Arm Microplate Handler, LiCONiC StoreX STX88 Incubator, and Hudson SOLO Liquid Handler – Performs liquid handling by diluting and mixing stock cell and antibiotic concentrations. E. coli was used as the stock bacterial cell and tetracycline was used as the antibiotic. E. coli's susceptibility to tetracycline serves as a proof-of-concept for the growth curve autonomous experiment. A Python script allows scientists to specify their desired experiment by uploading an Excel file designating which antibiotic and cell columns should be used for the creation of each assay microplate. This growth curve experiment has been proven to produce up to 12 assay microplates without significant human intervention. The Workflow Execution Interface (WEI) server acts as a link between the human-interactable portion of the application and the robot actions. The Python script connects to the WEI server and sends it the corresponding workflow files. The WEI server receives this workflow, and a worker object calls each robot action sequentially by sending a message to the robot node. The robot node translates the valid message from the WEI server into a form that can be executed by the robot’s lower-level driver, a series of functions to control the robot, and broadcasts the robot’s state (Ready, Idle, Busy, etc.) to the WEI server to indicate how the implementation of the task is occurring.
Results, Conclusions, and Discussions: When the T12 E. coli growth reading was measured in combination with varying tetracycline antibiotic concentrations, the average blank-adjusted optical density measurement at ½ the initial tetracycline concentration was 0.19 ± 0.01, 0.28 ± 0.03 at ¼ the initial tetracycline concentration, 0.46 ± 0.02 at 1/8 the initial tetracycline concentration, 0.81 ± 0.06 at 1/16 the initial tetracycline concentration, 1.11 ± 0.04 at 1/32 the initial tetracycline concentration, and 1.33 ± 0.03 with no treated antibiotic. There is a negative logarithmic relationship between growth activity defined by optical density and antibiotic concentration, with an R2 value of 0.951. This negative trend in growth demonstrates that the E. coli cell strain is susceptible to the tetracycline antibiotic, which supports previous literature. The maximum optical density reading of a blank well is 0.003, which is 1.7% of the smallest measured optical density reading, 0.173. This indicates that there is little to no contamination because of the robotic actions.
The finalized setup can accommodate up to 12 96 well plates in a single run, with the only human input needed from the start of the experiment to publishing the processed data on a global portal being setting up tip boxes. One 12 hour run of the setup matches the work of an entire lab working a day straight, and this approach does not lead to human error as all actions are pre-coded. The proof-of-concept growth assay between E. coli and tetracycline validates the experimental setup.
While the automated growth assay serves as a first step in validating autonomous biological experimentation, there remains much room for improvement. For example, the current protocol is limited to a maximum of 12 runs because the Robotic Arm Microplate Handler cannot reach positions on the Hudson Solo where stock cells and antibiotics would be switched. Additionally, more room would be needed to store and dispose laboratory materials such as tip boxes, 96 well plates, and 96 deep well plates. Finally, the setup can be made more affordable by synthesizing many of the different robots' actions into a singular Opentrons robot, allowing for a more streamlined and cost-efficient process.
Acknowledgements (Optional): Sincerest thanks to Abraham Stroka, Casey Stone, and Rafael Vescovi for guiding me throughout the internship and providing the opportunity to develop the multiple well plate experimental application. Truly grateful to mom and dad for inspiring my scientific journey.
