Practical considerations for the high-level automation of a biosciences research laboratory

.


Introduction
Currently, the execution of laboratory work in different domains of biosciences depends primarily on human operators, with varying levels of expertise.Familiarity with a specific work environment and with the task at hand renders it more or less laborious to the scientist.A critical factor to consider is that for most researchers, the time dedicated to experimental work represents only a fraction of their typical workday responsibilities.The complexity of time management challenges can be aggravated due to visible or non-visible disabilities, and can impose further complications with regards to the execution of experimental work [1,2].
Automated and robotic solutions have been developed to reduce such challenges as those mentioned above.Approximately 89 % [3] of current laboratory protocols have been analysed and rendered sufficiently feasible for splitting into smaller and individual tasks that can be easily completed using an automated equipment.More specifically, the majority of tasks in a bioscience laboratory rely on performing the manipulation of liquids.These tasks can be further expanded by linking analytical devices to liquid handling platforms to achieve rapid sample preparation, execution, and analysis, and, in return, obtain reliable results [4,5].
The adoption of an appropriately implemented liquid handling device may allow a researcher time away from the laboratory to focus on experimental planning, data analysis, and other tasks [4].It can empower experimentation by allowing the exploration of a larger design space via high-throughput screening [6][7][8].Moreover, the widespread adoption of automation using liquid handling devices, together with translation of current manual protocols into their automated counterparts, can reduce the impact of facility closures in pressing external circumstances, such as those experienced during the Covid-19 pandemic [9,10].It has been reported [11] that laboratory closures during the pandemic affected researchers by preventing access to their workplace and due to a widespread reduction of overall activities, affecting at times, even those that are not necessarily based at a laboratory.During the transition period, once institutions were back in operation, stricter social distancing and personal protection measures were put in place [12].Consequently, hybrid working, or shift patterns were adopted in many workplaces.Reflecting on that experience, strategically installed automated facilities or devices with appropriately trained staff and technical support would provide the optimal use of available resources to cause minimal disruptions.This could allow researchers to dedicate their time to plan experiments, then invest a short period of time in the laboratory to prepare samples, and, lastly, leave the execution of the experimental work to the liquid handling device.Strategically organised technical staff would prove invaluable to populate a device worktable and facilitate this process.This automated scenario is, in principle, similar to the workflow that occurs in a biofoundry [13].From the perspective of an individual research, looking into the future, the adoption of automation would also help to create a physical separation between experimental design and its execution.Ultimately, this will help to increase the quality of the results.
Despite the numerous advantages of suitably implementing an automated liquid handling device into a bioscience laboratory, technology adoption remains complicated due to numerous entry barriers.One of the major limiting factors for deploying a liquid handling device is its cost [4].The financial burden of a device can vary extensively depending on the brand, model, and the specific capabilities of the device [14].Devices, which can be tailored to specific applications, are available; some can include different modules, or others can work as multi-device platforms [14].All these efforts seek to reduce the capital cost required or to streamline the device specifications to create a well-focused expense item in the budget.
An additional entry barrier is the manner that research funding structures work, particularly in academic settings.Resources need to be managed in the most efficient way as they are a high-value asset to every academic institution.This created fixed-term and project-based research opportunities, particularly at the postdoctoral level [15].Postgraduate students are dedicated to experimentation only for the duration of their studies, and their research projects require the utilisation of a range of different experimental protocols, typically for limited periods.For "operators" with time limitation, it is important to ponder the cost-benefit aspects of learning any new skill, which would entail training, polishing, and execution elements.A researcher needs to decide at this stage whether automation will improve or interfere with the execution of their experiments, and consequently, their progress.Automation has been very successful in settings where tasks, once designed and programmed, remain the same for prolonged periods of time [4,5], but in an academic research environment, protocols need to change and adapt to accommodate discoveries and the requirements of not only different teams, but of individuals [16].This can pose a great challenge to adopt automation.On the other hand, using multiple protocol versions will cause a lack of standardization.This was observed in a recent study where protocol recording using electronic research notebooks was analysed.It was identified that researchers started using a template protocol but finished with 7 different versions, subsequently, the template needed to be modified to accommodate personal preferences and observations in the laboratory [16].In an automation setting, this would be a potential risk of burden to researchers as it translates into necessity for frequent modifications in their automated workflows.Unless this step is simplified and expedited, the researcher will make a conscious decision to avoid the utilisation of automated workflows due to the steep learning curve ahead and continue with manual work.
Another entry barrier emerges from the diversity of platforms available, at times, in a single institution or in a single operating unit of an institution.Each automated platform has a specific interface for users, which, despite possessing some similar characteristics, is practically different from those of other brands.Some require plain coding instructions; some can build automated workflows using complete graphical interfaces, while others are a combination of both.An understanding and experience in coding can considerably facilitate the implementation of automated protocols; nevertheless, learning to deploy a liquid handling device requires time investment to achieve flawless execution.Altogether, this contributes to the personal decisionmaking process of whether to implement this technology or not.
Such entry barriers need to be carefully evaluated and overcome in order to facilitate the integration of automation in research environments and thus expand the existing research practices.With this in mind, the implementation of a device-agnostic computational platform for the creation of workflows that does not require coding instructions can serve as a possible solution to overcome many, if not all, of the entry barriers.Such a platform would, first of all, remove the need to learn coding languages, and consequently allow the participation of a wider audience, with different backgrounds, to employ automation to enhance their experimental work.Circumventing the need to work through a long code could implicate speedy de novo workflow construction, and, for research-oriented applications, rapid workflow modification opportunities to accommodate continuously evolving experimental protocols.Lastly, a device-agnostic platform would mean that the same platform can be used for multiple liquid handling devices offered by even different brands.Altogether, this would mean that multiple academic or research units can pool resources to render them available to users across an institution, with minimal training requirements, without putting devices at risk of misuse.
The objective of the present work is to determine the suitability of a web-based, device-agnostic, and no-code platform (Synthace, Synthace Ltd.) to expedite and improve the adoption of automation in an academic environment where primary research activities involve the utilisation of a range of laboratory techniques adopted from biological sciences as well as engineering.The systematic evaluation was carried out by focusing on three components of the technology adoption process: 1) identifying the differences between the manual and automated operation rationale, 2) comparing the performance characteristics of manual and automated execution under ideal conditions (assisted execution), and 3) the evaluation of the practical adoptability of automation under real-life conditions (i.e., for the purpose of generating research data).Comparative evaluation of the manual and automated work was achieved first, by creating a flow diagram to visually represent the differences between these two types of experimentation.The processes were then investigated by comparing operational parameters such as the introduction of errors into the workflow, the time required for the execution of the task, and, where applicable, the repeatability of the analysis through the calculation of the This work is the first in its kind to demonstrate systematically the benefits of incorporating automation and testing a no-code platform in research environments with a large turnover of different users with varying levels of laboratory based practical and coding expertise and a wide range of different applications with relatively short project durations.Therefore, this work serves as a proof of concept to develop a guideline on how to deploy automation and enable its widespread use in an academic setting.

Liquid handling device
This work employed a Tecan Freedom EVO 200 ® (Tecan Group Ltd.) with a configuration depicted in Fig. 1.Briefly, the device is equipped with a liquid handling arm (LiHa) with 8 channels for disposable tips (DiTis) and an eccentric gripper arm (RoMa).Only the LiHa was utilized for experiments in the current work.Tecan devices have been reported to be one of the most used brands in research [14], and thus were deemed an appropriate platform to conduct the analysis.

Software
The workflow that a liquid handling device will follow needs to be programmed into a specialized software that will communicate with the device to facilitate the execution of the automated task.In this work, two software were employed: Tecan provides EVOware (Version 2.8.36.69 -Service Pack 4), which requires direct input of instructions for actions to be executed.EVOware works with a programming logic (instruction per line, inclusion of loops, and conditionals), but with a user-friendly dragand-place interface.It requires information regarding the manual input of the pipette channels to be used, the way the liquid will be manipulated (e.g., liquid classes: aspiration/dispense speeds, level detection, and delay times), the location to fetch the DiTis from the physical locations of plates, among others.
The no-code automation platform employed here is Synthace (Synthace Ltd., UK) (Versions 21.10.00 to 22.10.01).Synthace is a digital experiment platform that digitizes the entire experiment process from end to end.It helps scientists design and plan reproducible experiments, simulate them ahead of time, run them on their automated equipment, and automatically structure all their experimental data and metadata in a single platform.The Synthace workflow builder allows scientists to define their protocols using "elements", which express information about liquids or laboratory-specific actions, such as dilution, aliquoting, incubation, or mixing.These elements are connected as a diagram to express the protocol to be executed.The platform generates a preview of all the actions what will be executed, includes time estimates, and reagent usage.These instructions are then automatically converted to instructions to the liquid handler software, in this case, Tecan EVOware.
A representative comparison of EVOware and Synthace instructions for an aliquot process for 3 solutions and 3 replicates of each solution is shown in Fig. 2.

Experimentation under ideal conditions
To test the use of the LHD and the no-code software platform, a fundamental but comprehensive experiment was designed to perform a comparative analysis of a manually executed task and its automated counterpart that utilises a liquid handling device.One of the most commonly employed practices in every laboratory is cell passaging [17].This can be used for any organism from bacteria to mammalian cells growing on any type of culture media and can at times necessitate the incorporation of additional liquids and solutions as necessary.This experiment (cell passaging) was selected since a similar type of workflow can be employed to conduct other types of experiments including but not limited to the testing of cell survival, addition of different inducers, or preparation of cell samples for different incubation conditions, all of which rely on the mixing of cell cultures and other solutions at given ratios.
The designed experiment consisted of preparing a cell passage starting from a hypothetical previous experiment containing 10 cell culture samples (C i where i = 1,…10), the samples were to be tested for cell growth in 2 culture media (M i where i = 1,2), and challenged with 2 antibiotics (A i where i = 1, 2).Additionally, the cells were to be seeded in duplicates (r i where i = 1,2).This combination of variables will result in the preparation of 80 mixtures.As an arbitrary value, the total mixture volume was set to 170 µL with each component set at: C i : 10 µL, M i : 150 µL, A i : 10 µL.A representative diagram of this experiment is included in Fig. 3a.All liquids utilized for this experiment were represented as water since the aim is only to study experiment execution parameters, such as time for preparation, the required time for experiment execution, the amount of plastic material that is used, and incorporation of errors in the execution of the experiment.All these were recorded and analysed.
In the design explained here, additional factors that would be relevant in the context of an actual design to execute cell passaging experiments such as incubation times, the optical density measurements, or the sterilisation of materials were not included as they will impact both manual and automated execution similarly, and the primary focus of this experiment is to analyse experimental set-up and execution rather than what occurs before or after the preparation of the mixtures.
For manual execution, the experiment relied on three scientists from the department who agreed to conduct this experiment and the researchers had varying levels of expertise and familiarity with laboratory techniques: an MSc student working on their research project, a 2nd year PhD student, and a postdoctoral researcher.Instructions provided to the manual operators were as described above, and they were free to choose the type and amount of materials or consumables to use as long as the indicated design and volume requirements were satisfied.Briefly the researchers tasked to run the experiment manually were advised to treat this execution as any other experiment, which was critical to the interpretation of the performance.They were allowed to rest, be distracted, and to not isolate from other people in the laboratory, as distractions do  4) with a waste station (5), three separate plate carriers ( 6), an additional DiTi carrier (7), and a robotic gripper (RoMa).c) shows an aerial view the worktable and carriers available to be used for this study." is only shown when clicked on it once the "Define Liquids and Plates" is selected.Text boxes were included to explain each aspect of the workflow construction, numbers were added to the text to indicate the order of actions; text boxes with the same number are actions that can be performed simultaneously.b) shows the simulation result for the workflow presented in a), here the worktable, input plates, output plates, and additional tools are highlighted.c) shows the same workflow with same results created in Tecan's EVOware software.The worktable shows the same plate configuration as shown in b) and the script is labelled to identify each action performed.exist in a typical research environment.While they could serve as possible sources of error, they also help people relax, improve wellbeing, and reduce stress-associated errors.In these experiments, key metrics were recorded.The first metric was the time it took for the researchers to perform the preparation for the experiment, such as reading the instructions, making the necessary calculations, preparation of the solutions, and the retrieval of the necessary consumables.The second parameter was the time the actual execution of the experiment took: the preparation of the 80 mixtures on the microwell plate.Additionally, the researchers were asked to record the type and the number of consumables used in the experiment, during both preparation and execution.The same parameters were recorded for the automated version of the experiment and the results were compared (Fig. 6).As a final parameter of interest, the researchers were asked to make a note of the instances when they noticed any errors made during the experiment.This was used to detect the minimum number of errors for each run since the  possibility of errors gone unnoticed cannot be ruled out.This final aspect was not to be compared against automated experimentation, as that inherently assumes that no errors are introduced unless the workflow is built incorrectly.
For automated execution, a workflow was constructed in the no-code automation platform that replicated these instructions, and the details of this workflow on how to describe the preparation of materials, consumables, liquids required on the platform along with plate positioning in the liquid handling device's worktable are detailed in Section 2.5.

Experimentation under real conditions
The experiment described in the previous section was designed with the sole purpose of investigating the role of execution parameters.In this section, an additional experiment is described to evaluate the automated pipelines in an experiment using actual samples.For this authentic cell samples and culture media were used under aseptic conditions.The objective of this experiment was to construct a correlation between optical density (OD) measurements and colony forming units (CFU) for four different bacterial organisms from four different genera (Bacillus spp., Enterobacter spp., Pseudomonas spp., and Rhodococcus spp.).This experiment consisted of inoculating 25 mL LB media in 50 mL tubes from a cryostock (10 µL approximately) and incubating at 30 • C and 200 rpm.Once the final optical density (OD) reached 1.0 and above, 250 µL of the culture was then used to inoculate fresh 25 mL of LB media.Cultures were incubated for 12 h with 1 mL samples taken every 2 h and replenished with fresh LB.For all species, samples were normalized to an OD of 0.1 and then, subjected to serial dilutions to be spotted on LB agar plates (1:20, 1:400, 1:8000, 1:80,000, and 1:800,000).Serial dilutions were prepared either manually or using a liquid handling device.Then the CFU count was correlated to their respective OD.The comparison between manual and automated execution was performed by assessing the quality of the execution of the preparation of the dilutions; the parameters of interest were the CFU count, variability (coefficient of variation), and correlation with OD (R 2 coefficient) to conduct this evaluation.A representative diagram of this experiment is included in Fig. 3b.

Automated workflow construction
For automated execution, a no-code workflow was constructed that replicated the instructions used by the researchers.Details for this implementation are provided in next subsections.The instructions on the amount of consumables used, types and volumes of liquids needed, and plate positioning in the worktable were provided by the simulation capabilities of Synthace, reducing variability in automated work.In the following subsections, the construction of these workflows will be described in detail to provide insight as to how the no-code automation rationale works and how a hypothetical demonstration would differ from actual experimentation.

Cell passaging experiment (Experiment Under Ideal Conditions)
Synthace provides code-free instructions to perform a set of actions.Instead of individual robotic actions, holistic experimental concepts, such as "dilution" or "aliquot", can be incorporated into elements as they are already defined in the software.Thus, it only requires an input to define what will be used in the experiment for the cell passaging protocol.First, the "Define Liquids and Plates" element was used to define the hypothetical solutions: 10 cells sample solutions each containing 200 µL (C i ) in a 96-well plate (Cell Plate), 2 antibiotic samples of 10 mL volume each in a 24-deep well plate (Antibiotic Plate), and 2 different types of culture media each stored in a 100 mL trough.
There are several options to define how the mixture are to be prepared, but the most direct and concise option is to employ mixture preparations with the combination of two elements: "Upload Mix Set Plan" and "Mix Set".The first element requires the upload of a table containing the order and volume of the solutions to be prepared (Supplementary Information 1).The order of mixture preparation is organized by layers, for this experiment layers 1, 2, and 3 represent culture media, antibiotics, and cell samples, respectively.Additionally, "Upload Mix Set Plan" requires the number of replicates, which for this experiment was set at 2. The "Mix Set" element is a pre-defined element for the execution of mixing of the liquids.The only input required from the user is to define the destination plate.In this instance, the destination plate was defined as a 96-well plate as this is the most common plate type that can be used for incubation and for OD600 measurements using a plate reader.This plate was defined as "Mixture Plate" and Synthace will allocate the defined mixture there.
An additional feature in the platform is the option to create "Templates".This is a workflow format where a user creates a specific workflow, but for subsequent runs the experiment sequence cannot be modified except for specific input variables determined by the creator.A Template was constructed for the cell passaging experiment where the only input variables allowed to be changed were the definition of the liquids to be used, the master plan for the mixtures (layers and volumes), total mixture volume, and the number of replicates.A screenshot of this template is included in Supplementary Information 1.

Serial dilution experiment (Experiment Under Real Conditions)
For the serial dilution experiment explained in Section 2.4, a workflow was designed to perform serial dilutions within a specific range of final concentrations (1:20, 1:400, 1:8000, 1:80,000, and 1:800,000).The cell cultures were performed for one specie at a time, and the dilutions were prepared for the biological triplicates and for a negative control (LB-only culture).In this workflow, a key element called "Dilute" was used; this element encompasses all necessary actions during the preparation of a dilution protocol including liquid to dilute, dilutant, volume of dilutions, and final concentrations.In the following paragraph, we provide a detailed representation of how this workflow is created in order to highlight the differences in the thought process that a researcher adopts when defining their experimental workflow in the nocode environment in comparison to how the typical workflow would be executed if the experiments were conducted manually.
To start the automated workflow, two copies of the "Define Liquids and Plates" element were created.One of them was used to describe the cell culture samples provided in triplicates and previously normalised to 0.1 OD along with the negative control of the experiment.The second element was used to define the diluent, which in this case was a PBS solution.The next element defined was called "Dilute", which is the element in charge of controlling the calculations and the preparation of the dilutions.This element requires several inputs.First, the dilutions are to be prepared in a "New Location" meaning that they will be prepared in a new well and plate.The task of dilution will also be executed in a serial manner with the following dilution to be prepared using the preceding one at a pre-defined concentration or dilution factor.Once these are defined, the diluent for each sample can be specified.Dilution volume was fixed at 2.5 mL including those of intermediate dilutions.The target concentrations are entered for each of the 5 dilutions.This step offers some flexibility to define dilution using different units and formats, but for ease of operation the "Amount of Times Diluted" operation was adopted.For these experiments the first 3 dilutions utilised a 20-times dilution (5 %) and the last 2 used 10-time dilutions (10 %).Considering that all started with the undiluted well concentration X, this led to 0.05X, 0.0025X, 0.000125X, 0.0000125X, and 0.00000125X in consecutive wells.As the final step, a target plate was defined and it was placed as a 24-deep well plate, on which only 20 wells would be occupied.An additional element "Aliquot" was introduced to facilitate OD measurement by a plate reader by collecting 100 µL from each of the dilutions, which would then be placed into a new 96-well plate.This final step is not absolutely necessary, as individual dilutions could also be measured using a spectrophotometer and UV-Vis cuvettes.
Screenshots for both automated workflows are included in the M.A. Torres-Acosta et al.

Comparison of manual and automated workflow construction and execution
The comparison between the manual and the automated workflows was carried out by considering the construction of the work plans and, their execution separately.For the dilution experiment detailed in Section 2.4, an additional comparison was made using the actual experimental results.
The manual and automated works were initially analysed through the construction of flow diagrams that attempt to replicate the thought processes for each option.This evaluation included how the inputs into the process such as the set of samples, plates, pipette types and tips and other consumables would be allocated, how the calculations for well or plate volumes are done, and how the liquid transfer operations are conducted.This comparison was then expanded to include a comparison of the no-code automation with that of the standard software of the liquid handling platform.The emphasis in this comparison was to compare two automated workflows, that of Synthace to the autogenerated script in EVOware by connecting the script to Synthace for the actual execution.
The execution performance for manual and automated experimentation was compared using the preparation time of the samples, the actual execution of the cell passaging experiment, and the errors identified in manual execution as the primary evaluation metrics.The extent of consumable usage was compared across the three lab operators with varying levels of expertise in laboratory practice and the automated execution.Synthace incorporates the term "liquid policies", which stands for rules or suggestions a user can make to the platform on how liquids should be handled, these policies can be defined in the platform and are matched to specific liquid classes.These last are how a liquid is actually handled on the liquid handling device.Through this mechanism, different liquid policies were tested to analyse how consumable usage could change by altering how the antibiotic and cell samples are handled in order to try to reuse tips.The already defined and ready-to-Fig.4. Flow diagrams that represent the logical workflow for designing the experiment as employed by a human operator and as executed manually (a) and by an automated platform (b) for a cell passaging experiment.Flow diagrams include variables that are calculated to determine the amount of labware required.
use "DispenseAboveLiquidMulti" liquid policiy allowed the simulation of an experiment prior to execution and the recording of consumable usage allowed such comparisons to be made.
As a final performance test, individuals with no prior training on the platform or at times, even the liquid handling device, were challenged by providing them with a pre-designed template for the cell passaging workflow as described in Section 2.5.1.This helped to obtain verbal feedback on their non-guided experience to use the platform.Lastly, as mentioned before, for the serial dilution experiment, additional parameters, such as coefficient of variation and the correlation coefficient (R 2 ) were analysed for the regression equation of the form: CFU (mL − 1 )= β 0 + β 1 × OD + β 2 × OD 2 where β 0 was set to zero.

Flow diagram construction
A flow diagram is a very useful tool to visualize processes and becomes invaluable when designing an experimental protocol as the order of the steps is easily affected by the experimental tools or devices employed.In this study, the contrast between manual and automated work has generated differences in how the same experiment is executed.As mentioned in Section 2.6, the experiment that comprises the preparation of mixtures of cell samples, media, and antibiotics in the form of a flow diagram was compiled in Fig. 4 for the manual and automated versions.
Using these diagrams, it is possible to visualise that during an automated workflow some steps can occur simultaneously.For example, inputs of chemicals, data, and consumables can happen at the same time, while in its manual counterpart it tends to be in more linear fashion.Particularly for this experiment, the initial inputs are the data for cell samples, the type of culture media to be used, and the availability of additional solutions, such as the antibiotic solutions proposed.Then, calculations are made to determine the amount of liquid to be moved for each operation and, subsequently, solutions are physically brought together to create the mixtures (Fig. 4a).For the automated work (Fig. 4b), the data, the calculations required, and the solutions definition need to be presented since the beginning and then, the workflow operation can proceed without delay and by itself.
A critical aspect to note is that overall, both workflows are comprised of 16 operations including the inputs, decisions to be made, and the outputs, but in the manual work, the linear manner of work renders all 16 steps imperative to be followed subsequently in order to obtain the final working plate.The automated alternative can be considered as six operations shorter due to the simultaneous nature of the processing of the inputs.However, similar tasks need to be executed in different ways; an automated workflow requires interaction with the liquid handling device to populate the respective worktable and be physically ready to operate, while manual work is conventionally experimental.The experiment designed here would be able to accommodate a degree of ambiguity depending on the requirements for the final outcome of the experiment by every user unlike earlier reports [18].For this reason, the flow diagrams do not show specific volumes or sample numbers.As this experiment is intended to support academic research, the necessity for accommodating user-specific protocol requirements is expected to be more prominent leading to more ambiguously defined workflows and templates.

Comparing manual and automated execution
During the execution of a workflow, the no-code platform connects with EVOware and generates a specific script that will perform the programmed task.As the platform can communicate with different LHD brands, the auto generated scripts are not as succinct as those generated directly on the software by trained expert users.This does not compromise on efficiency and will perform similar to an optimally encoded script.To compare the execution performance, the experiment under ideal conditions, for which the goal was to prepare 80 different mixtures, was implemented in a no-code workflow and the automatically generated script was further investigated.The ability of Synthace not to compromise on efficiency when communicating with the Tecan demonstrated its strength to effectively reduce the burden of programming by its no-code approach.For both automation platforms, the workflow construction was evaluated under the assumption that users with minimal coding experience would attempt to automate their protocols in the academic research setting and thus the automatically generated scripts would serve as a good representative of non-optimal scripts mimicking a typical output from a user who received limited training on EVOware.The script is heavily populated with repeats of simple commands for aspiration, dispense and wash of tips.An example of such a protocol developed in Synthace and the EVOware-converted scripts are provided in Fig. 5, which allows a comparative evaluation of the nature of both platforms to facilitate automation.The no-code workflow is essentially designed as a flow diagram of the task to be performed, and EVOware's script is constructed using a coding logic containing a series of instructions ordered by lines, with every action to be programmed.
The script shown here could be shortened extensively through the introduction of loop instructions in this experimental setup under investigation, but it is important to consider that in a real-life scenario, every single mixture could be different, and, in turn, a very complex and long script would then be required.In such cases, any error made in programming could easily be overlooked and introduced into the experimental protocol.
Following this evaluation of the different automation workflow building platforms, the performance of experimentation conducted through automated workflows was compared to those conducted through manual execution by colleagues in the laboratory.Researchers were tasked with the same instructions presented in Section 2.5.1 and were given a document containing specific volumes and setting up of the experiment (file can be found in Supplementary Information 2).
Automated work was conducted using the fixed tip option (no disposable tips used) for the Tecan liquid handling device and also the DiTi option (use of disposable tips) with the incorporation of different liquid classes, namely (i) assigning no specific liquid classes (Synthace auto-selected a base liquid class), (ii) dispensing the antibiotic solutions above the liquid level, or (iii) dispensing the antibiotic solutions and the cell suspensions above the liquid level.Depending on the liquid policies applied, the time required for completion of the experiment varied (Fig. 6a).This difference was caused by the number of robotic movements the device needed to perform to fetch and dispose DiTis.This did not apply to the fixed tip Tecan platform as it required a tip wash between each liquid transfer to remove remnants of previous liquids.
Time requirements were shown to vary substantially, even for such a short experiment as the one described here (Fig. 6a).The time required for automated execution in a Fixed-Tip device was approximately 10 min for preparation and 15 min for execution, while the manual work took up to three time that of its automated counterpart.This was identified as a key difference between manual execution and robotic deployment, which renders the utilisation of automation an option worth deliberate consideration to maximize a workday schedule.One striking advantage of utilising automated platforms was the removal of experimental variability and consequently the minimisation of uncertainty.It was inherently accepted that automated platforms displayed minimal variability in the time it took to execute the technical repeats of the same experiment.On the other hand, the extent of variability in the time parameter across researchers was striking (Fig. 6a).As expected, an automated execution will not change between runs as it always behaves in the same manner.Contrarily, manual actions can be carried out in different ways.The error bars presented are the standard deviation of the execution time when the experiment was conducted by different users, which relates to their level of expertise and familiarity with the laboratory.The variation proposed on the automated work analysed here was introduced to determine whether a single device, performing the same actions, would take different execution times.
One important advantage of the no-code platform was the option to perform simulations of the experiment prior to the actual execution.This simulation provided a snapshot of the required configuration of the worktable including the placement of consumables -DiTi and plates and the required volumes of input solutions and samples.In this case, the platform could advise the user to determine the minimum volumes of input liquids and solutions in the event that the input volumes were not defined in the "Define Liquids and Plates" element.These simulations can advise the user on the necessary worktable adjustments to ensure a smooth automated execution.It also advises on the extent of consumable use and can assist with the minimisation of single-use plastic waste.In this setup, this feature was used to compare the extent of consumable usage by the researchers with that of the automated platform, in either one of the four different automated executions: fixed-tip Tecan platform, DiTi Tecan platform with the default liquid policies, DiTi Tecan platform with liquid policies defined for antibiotics, or for both antibiotics and cell samples (Fig. 6b).
The type of liquid transfer strategy (a Fixed-tip or a DiTi Tecan device) will determine the extent of requirement for disposable tips.However, the way starting solutions are handled and prepared remains the same regardless of the liquid transfer strategy employed.In this experiment, all solutions were prepared manually prior to their placement on the worktop, thus rendering the consumable usage for this step the same across different platforms and liquid transfer strategies, and the extent of human interference with the experimentation.The plates containing the starting solutions and the destination plate, i.e., the mixture plate, also remains unchanged across automated experiments indicating that, among different options tested, a Fixed-tip device It also shows the device configuration, available tips, and labware to use.b) shows the complete EVOware script auto generated by Synthace to perform the same tasks.This was used as representative of a user with very basic coding skills that would lack an understanding of the use of loops and conditionals typical of programming.This is used to mimic the experience of a user untrained in programming, who has access to the software.requires the minimum number of consumables for the execution of the liquid transfer (Fig. 6b).Not all applications are suitable to adopt such an approach, and therefore the utilisation of disposable tips and the assignment of different liquid classes was investigated for such applications.The benchmark for tip usage was selected as the Fixed-Tip operation, where only 2 × 1 mL tips, 12 × 200 µL tips, 3 × 96-well plates, and a 24-deep well plate totalling up to 210 g of plastic were needed for preparation and execution.The other extreme of the spectrum, where no liquid classes were employed, required the same amount for preparation, but for execution it relied on additional 16 × 200 µL tips and 168 × 50 µL tips, accounting for an increase of 66 g (for a total of 276 g) of plastic waste as a consequence of only using each DiTi once.The optimal usage of disposable tips on a liquid handling device is only possible when tips can be re-used, particularly when dispensing the same solution multiple times.This is a typical but risky manual practice, too, which is frequently observed for lab practitioners, but the elimination of the possibility of cross-contamination between wells lies solely on the experience of the experimenter and it may introduce unforeseen errors into experimentation, of a type that is not quantifiable by selfreporting in most instances.In contrast, in an automated system, this problem is solved by ensuring that the liquid or the solution was dispensed without touching the meniscus in the destination plate wells, thus avoiding any possible source of contamination.The material use was reduced in this manner by 104 × 50 µL tips, which corresponded to a total 240 g of plastic waste for this simple experiment.Plastic consumption was observed to vary extensively during manual execution by researchers.The protocol for the experiment was left semi-ambiguous precisely to be able to demonstrate such variability.In response to this, some researchers were observed to employ larger screw-cap tubes as containers and serological pipettes for moving larger volumes of liquid (Fig. 6c).On average, the plastic waste generated during manual experimentation was determined as 121 ± 26 g for this experiment (Plastic waste generation data included in Fig. 6a).
A clear advantage of this no-code platform was the ease with which modifications can be introduced to a workflow by defining broad and general actions.The encoding requirements of the full-code platform would require, instead, the definition of liquid classes as unique steps for each individual step in which any type of liquid was manipulated through the aspiration and dispense actions.Furthermore, if the same liquid was to be manipulated multiple times, but in separate lines of command, then this change needed to be represented every time the liquid was explicitly mentioned in the script.This is further detailed in Fig. 5.In the same manner, if different volumes for a new set of mixtures are to be used, the changes in EVOware need to be performed separately for each transfer (i.e., for each aspiration/dispense cycle in this case).It is possible to add equations, conditionals, and loops for ease of coding, but if every scenario is different, then each individual input will be different and unrelated.Using Synthace, the preparation of a spreadsheet file to upload is far easier and most researchers have experience using spreadsheets, such as those tools within Microsoft Office packages (Microsoft Corporation, USA).
An additional consideration concerned the errors introduced by manual operation.All researchers reported at least one error and the maximum was reported as three errors, indicating that a simple, but laborious task, such as this pipetting protocol, can be prone to up to 4 % or even higher rates of non-technical errors.It is important to note here that the errors mentioned were those that the researchers were conscious of and does not represent the complete picture.Automation ensured the prevention of such non-technical errors by eliminating their possibility entirely.Considering the duration of the experiment and the errors encountered as reported by the researchers totalling 1.667 errors per execution on average, it can be concluded that an error was expected every 23.8 min.Expanding this to a more complex experiment, for example if three replicates of each experimental condition were considered for experimental design instead of the two replicates used here, the length of the experiment would substantially increase when executed manually, and consequently additional errors would have been introduced along the way.Altogether, execution errors are inevitable, and they cause large delays in the completion of the experiment.Automation was shown to provide an irrefutable solution to prevent them in this scenario.

Experiment in real-life conditions and comparison with manual operation
A second experiment was designed where actual cell samples and culture media were employed.The comparison focussed on the execution time, the material and consumable requirements, and the quality of the results obtained.An important factor to consider was that the researcher who assumed the role of operator for this experiment was not trained to use the robotic device nor its software.Instead, the user received minimal training in using the no-code automation platform with the aim of understanding how to schedule an execution.
The performance measures for the automated and manual operation are summarised in Table 1.Execution time was observed to differ extensively.The total execution time for diluent addition and dilution preparation required 25.5 min of manual input, while the automated execution only needed 12.7 min.This simple experiment was sufficient to highlight that even for simple and short experiments, such as in dilutions preparation, the execution time can still be optimised through appropriate implementation of automation.In this experiment, manual operation generated 37.4 g of plastic waste, while the total plastic waste was 31.5 g for automated operation.This corresponds to only 16 × 1 mL and 32 × 200 µL tips for automated operation and 20 × 1 mL tips + 1 × 25 mL serological pipette to prepare each dilution and transfer them into a new 96-well plate or 20 UV-Vis cuvettes for optical density measurements.It is important to mention that the duration of the work and the plastic consumption noted above are reported for the preparation of the serial dilutions required for the execution of the experiment for a single organism out of the 4 different species investigated in this experimental setup and for a single time point out of the 5 time points at which samples were collected.So, the differences in the duration of the complete experiment executed by a researcher or through an automated platform will proportionately reflect on the number of errors introduced as we discussed earlier.
The data acquired from manual experimentation and automated workflows were investigated in three aspects: prevention of contamination, the correlation between OD values and CFU counts when different modes of experimental execution were employed, and the variability across replicate dilutions.Since actual culture medium was used in the experiments, ensuring contamination-free execution was of utmost importance.While the liquid handling device employed in this study was not designed to operate in an aseptic environment, the operating conditions were maintained contamination-free by treating the worktable with ethanol.Movable parts(i.e., the LiHa) were cleaned using lint-free tissue sprayed with ethanol and swiping very carefully not to remove any lubrication allowing uninterrupted mobility for the device.All dilutions were accompanied by LB negative control to monitor any possible contamination issues.None of the negative controls were contaminated in the preparation of 20 dilutions for each of the five time points and four different species.In contrast, contamination was observed to be an issue during manual handling of the samples due to the possibility of cross contamination or lack of focus during extended working hours.
Following the preparation of dilutions, each one was plated onto LB agar and incubated as described before.The CFU counts for different dilutions were determined considering only those counts between 30 and 300 CFUs as described earlier [19,20].These CFU counts and OD values were then used to determine the linear correlation equation and the goodness of fit of the correlation model (Table 1).While the goodness of fit (R 2 ) was acceptable for all manual execution, the use of a liquid handling device improved the fit in all cases.This is the result of standardisation enabled by the way the liquids are transferred, always obeying the same rules on speed, tip position, sustained mixing, among others.The coefficient of variation for the results of the experiments coming out of an automated workflow was much lower (up to 3.5 time lower) than those obtained from manual experimentation in all instances providing good insight on repeatability of experiments.Similar to the case of the 'hypothetical experiment' presented in Section 3.1.2,errors were introduced during manual operation, which rendered the experiment incomplete as opposed to its automated counterpart.

User feedback on automated work deployment
After researchers completed their tasks executed by manual operation in the lab, they were asked to investigate and manipulate a low-code cell passaging template, which replicated their manual task, with the aim to understand their perception of using an automated device with a no-code software.The template was specifically designed to prevent any possibility of modification to add or delete steps and the task to be executed was 'fixed' in that sense to what was initially programmed.The

Table 1
Results obtained for the correlation of colony forming units per mL (CFU mL − 1 ) and optical density (OD) at 600 nm.Results included are regression equation coefficients (β1 and β2), the correlation coefficient (R 2 ), the coefficient of variation (CoV) and the consumables used for each type of execution.only modifications that were allowed by the programmer of the template were the liquids to be used, the master plan for the mixtures (order of addition and volumes), the number of replicates, and the final volume of each mixture (Supplementary Material 1).The most common inquiries were related to typical challenges encountered whenever a new device or software is deployed in a research environment.Users were reluctant to explore the interface.Options in the user software, which would enable a more detailed view of the workflow and the different elements coded into it, passed unnoticed and unexplored by the minimally trained users.Furthermore, the users were reluctant to resort to interactive help available on the template interface.As a final point, all users were hesitant to modify and perform simulations with an explicitly expressed concern to sever the operation and damage the platform.The strength of the platform via its ability to run the simulations of an experiment prior to execution, and via its interactive help chat should be stressed further to overcome this type of reluctance, which was characteristically unexpected in young researchers who are generally very much open to exploration of novel venues.
Following this feedback session, a demonstration was made for the users to observe the result of their input with a real execution.All users were interested in the LHD execution and were impressed by the simplicity of the process.Nevertheless, the tool has already facilitated third party researchers' experimental work as demonstrated in Section 3.2, with the results and data obtained from that work being incorporated into the research project for a Masters-level program.

Conclusions
Robotic laboratory devices that facilitate the automated execution of experiments in biotechnology and biology laboratories often require the implementation of protocols to execute well-defined tasks.These protocols are often implemented by experts in the utilisation of the software of the platform who also have some level of programming or coding experience.The researchers then utilised these pre-packaged protocols in their day-to-day experimental activities.This division of labour works very well under professional settings where a similar type of experiment with minimal intervention to the protocol except for modifying parameter ranges is repeatedly employed.On the other hand, a different type of challenge emerges in a research environment where novel experimental designs are required by different individuals to execute different tasks for relatively limited periods of time.The value, need, and the necessity to learn how to programme such a device appropriately to automate the task at hand may implicate a critical trade-off for the researcher in terms of allocating their resources.Consequently, alternatives that forgo the need to allocate time to learn how to use and programme a specific robotic device are valuable options to explore such settings.Furthermore, the utilisation of automated platforms for experimentation is especially attractive to ensure high quality data generation with minimal use of resources regardless of the experience of the researcher, which is inherently quite variable in an academic setting.This final point renders the exploration of practical alternatives to overcome the challenges stated above non-futile.
In this work, the suitability of automation platforms that require no coding experience or knowledge, i.e., no-code platforms, in an academic research environment was investigated systematically.We demonstrated here how the incorporation and testing of a no-code automation platform helped to transform simple protocols into automated tasks, which were executed faster in an error-free manner to render work more efficient than its manual counterpart.Furthermore, the relatively straightforward adoption of the technology by the researchers with little to no coding experience nor any training on the use of the robotic devices indicated that no-code automation platform offered a viable and resource-efficient solution to integrate robotic operation into bioscience and bioengineering research in this academic environment.The platform used here, Synthace, allowed the users to evaluate their consumable requirements through simulations of an experiment prior to execution, and offered alternatives to reduce plastic waste through the editing of liquid policies.Users were able to create multiple scenarios, and swiftly modify them to find the optimal outcome for their experimental design without the necessity to commit to an actual experiment and waste valuable samples and other resources.
The adoption of the no-code platform was shown to be expedited to the service of partially trained users in this study.This was shown to be especially beneficial in the serial dilution of cell suspension experiment, which was rendered highly sensitive to human error, inter-repeat variability and contamination as shown by the CFU outputs, and the adoption of the automated workflow removed an important barrier in laboratory work skill especially for early-career researchers with limited experience working in a laboratory.The initial learning curve for the adoption of the no-code platform was observed to be very gentle even for untrained users offering a viable alternative to fully manual execution of day-to-day laboratory experiments.We believe that this work can serve as a guideline on how to approach the transition into automated experimentation and using simple, every day, protocols demonstrate to users and principal investigators how automation can be incorporated into laboratories to maximize the potential of these devices.

CRediT authorship contribution statement
MAT-A and DD conceived the study and experimental set-up.MAT-A, FCR-C and PB performed experimental work.MAT-A performed all automation work and wrote the manuscript.DD, NL, RK, and GL provided critical revisions, formatted, and edited the text, and approved the final version of the manuscript.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Fig. 1 .
Fig. 1.Configuration of the Tecan Freedom EVO 200 used to perform all the work in this study.a) shows the overall device, b) shows a detailed image of the front of the worktable with numbers highlighting each component: liquid handling arm (LiHa) (1), tip wash station (2), carriers for troughs (3), disposable tips (DiTi) carriers (4) with a waste station (5), three separate plate carriers (6), an additional DiTi carrier(7), and a robotic gripper (RoMa).c) shows an aerial view the worktable and carriers available to be used for this study.

Fig. 2 .
Fig.2.Typical Synthace workflow and EVOware script for an aliquot process for 3 samples with 3 replicates for each solution with 9 wells used in total.a) shows the main screen for Synthace to construct workflows; this includes the menus displayed only when clicked on the elements.Additionally, the "Liquids Table" is included underneath the flow diagram to show how the liquid definition looks.The "Liquids Table" is only shown when clicked on it once the "Define Liquids and Plates" is selected.Text boxes were included to explain each aspect of the workflow construction, numbers were added to the text to indicate the order of actions; text boxes with the same number are actions that can be performed simultaneously.b) shows the simulation result for the workflow presented in a), here the worktable, input plates, output plates, and additional tools are highlighted.c) shows the same workflow with same results created in Tecan's EVOware software.The worktable shows the same plate configuration as shown in b) and the script is labelled to identify each action performed.

Fig. 3 .
Fig. 3. Schematic representation of the experiments used to test the implementation of automation.a) shows the experiment that was performed by the automated platform (Tecan Freedom EVO 200) and the manual users.Only water was used to replace different types of solutions, but it was treated as real samples in different wells.This facilitated the analysis to study use of automation rather than actual experimental results.b) shows the execution of an authentic experiment used to test implementation of automation and compare the results with those from the manually executed experiments.Automation was incorporated in the preparation of the serial dilutions to determine a correlation between optical density (OD) and colony forming units (CFU).Figure developed in Biorender.
Fig. 3. Schematic representation of the experiments used to test the implementation of automation.a) shows the experiment that was performed by the automated platform (Tecan Freedom EVO 200) and the manual users.Only water was used to replace different types of solutions, but it was treated as real samples in different wells.This facilitated the analysis to study use of automation rather than actual experimental results.b) shows the execution of an authentic experiment used to test implementation of automation and compare the results with those from the manually executed experiments.Automation was incorporated in the preparation of the serial dilutions to determine a correlation between optical density (OD) and colony forming units (CFU).Figure developed in Biorender.

Fig. 5 .
Fig. 5. Comparing the cell passage experiment in Synthace (a) and EVOware (b).a) includes all the cognate menus that are displayed when each element is selected.It also shows the device configuration, available tips, and labware to use.b) shows the complete EVOware script auto generated by Synthace to perform the same tasks.This was used as representative of a user with very basic coding skills that would lack an understanding of the use of loops and conditionals typical of programming.This is used to mimic the experience of a user untrained in programming, who has access to the software.

Fig. 6 .
Fig. 6.Performance evaluation and comparison between the manually executed and automated cell passaging protocol.a) shows the comparison between the preparation and execution time for manual work, the use of Fixed-tip Tecan platform, the use of a DiTi Tecan with no tip re-use, the use of a DiTi Tecan with tip re-use for antibiotics (denoted as Automated -DiTi Re-Use 1), and the use of a DiTi Tecan with tip re-use for antibiotics and for cell samples (denoted as Automated -DiTi Re-Use 2).The error bars denote the variability across technical repeats of the experiment using an automated platform, or across different researchers during manual execution studies (n = 3 in each case).b) shows the material usage for all automated versions with clear delimitations for each liquid class consideration.c) shows the diverse material usage of researchers who handled the experiments manually.