HORIZON DISCOVERY
Application of Petri Nets to Laboratory Automation Scheduling Peter Grossman and Anatoly Myaskovsky Horizon Discovery, 245 First Street, Third Floor, Cambridge, MA 02142, U.S.A
Workflow Overview We examined alternative scheduling solutions for laboratory automation. Considerations included predictable order of actions, ease of configuration change and optimized use of instrumentation. Tradeoffs across these parameters were required. Our needs dictated that we prioritize throughput over predictable behavior. A scheduler which exhibits simplicity and transparency (both qualitative terms in this context) demonstrates greater advantage in a performance centric or changing environment. However, we selected a system which emphasized a non user‐directed order of actions for multiple iterations of a protocol to enhance throughput. Building this solution “from the ground up” has allowed us to evaluate non‐ traditional component solutions.
Introduction At Horizon Discovery we process tens of thousands of microtiter plates per year in support of our cell based screening operation and its library of thousands of commercially available and proprietary cell lines (Figure 1). The Technology Group is tasked to identify and deploy automation solutions when throughput is a priority and protocols are understood to a level at which automated solutions may be applied.
Colored Petri Nets
Strict Timing Control
Colored petri nets support the customization of tokens. In our case, tokens are a stand‐in for assay plates, each with specific properties (e.g., cell type, density). Standardizing these properties along with the use of intelligent incubators allows the model to personalize the priority and routing of plates.
User configuration of the model supports staggering the entry of plates, limiting the capacity of hotels and customized delays at transitions. Capturing timings is replayed in simulations in accelerated time scales.
Results In our application we chose to represent physical actions (i.e., plate/robotic arm movements) as transitions and physical locations and instrument operations (or their surrogates) as places (Figure 3). In the application below, Petri Nets are used to increase the efficiency whereby reagent is added to thousands of microtiter plates prior to plate disposal (Figure 4). Transition firing based solely on token availability maximizes availability of the robotic arm.
Standards Based Components There are numerous open source implementations of petri nets as well as many articles on their theory and use. Our driver implementation uses TCP, Serial and USB connections to modern platforms (Windows, Linux, even Android) supporting web based access. We leverage open vendor APIs to enhance instrument performance such as confirming arm motions, delivering files and monitoring bulk dispenser performance with a USB camera (Figure 5):
Response Area 10.0 7.5 5.0 2.5 Figure 3: Reagent Add Protocol‐ Working Petri Net Example
0.0
Figure 1: Response Area of Oncology Focused Drugs Across a Broad Cell Line Panel. X‐Axis represents 290 oncology cell lines and Y‐Axis represents 75 oncology focused standard of care and emerging therapeutics.
Figure 4: Benchtop with Robotic Arm, Hotels, Dispenser, Incubator and Readers
Methods A Petri Net is a directed graph with an initial state called the initial marking. There are two kinds of nodes: places (i.e., conditions, signified by circles) and transitions (i.e., events that may occur, signified by boxes or bars) (Figure 2). Arcs occur from either a place to a transition or a transition to a place. Petri Nets have an exact mathematical definition of their execution semantics, with a well‐developed mathematical theory for protocol analysis. In practice, petri net schedulers typically have hundreds or even thousands of nodes. The strict mathematical definition of petri nets supports the designer’s ability to break up a large graph into many smaller subnets. These subnets may then be more easily managed and modified with a well understood impact on the remainder of the model. Figure 2: Petri Net. (Murata, T. Proceeding of the IEEE 1989 77(4) 543. Petri nets: properties, analysis and applications.)
t + 44 (0)1223 655580 f + 44 (0)1223 655581 e
[email protected] w www.horizonbioproduction.com Horizon Discovery, 7100 Cambridge Research Park, Waterbeach, Cambridge, CB25 9TL, United Kingdom
Scalability We chose to model multiple instrument copies behind each petri net place. instruments may have unique characteristics that are visible to the scheduler.
Individual
Protocol Flexibility When a transition fires it may have a choice among incoming tokens and outgoing places. The scheduler can look at the color of these tokens and choose among them according to performance needs or token characteristics. Likewise, the scheduler can choose among instances of an outgoing place as the target for a particular token.
Figure 5: Bulk Dispenser Time Lapse Monitoring With Density Analysis (Combi Photo with Time Lapse of Focus Area)
Conclusion Automation, in its own right, has inherent advantages in processing laboratory experiments. Utilizing a petri net scheduler has lowered the barrier to supporting increased production demand as well as handling workflow modifications in a predictable manner. We believe that this solution enforces a cleaner separation of workflow and instrument operation with the benefit of fewer side effects, better fault isolation and smaller time to recovery.