March 15, 2008 (Vol. 28, No. 6)

New Procedure Allows Monitoring Liquid Dispensing to 1,536 Microtiter Plates

Machine vision is the art of using computer-controlled cameras to maintain high levels of quality control in industrial processes. In manufacturing, parts produced are often automatically imaged and analyzed to determine if the product produced is acceptable. Often, feedback is associated with this procedure such that if an unacceptable part is encountered it may be automatically removed from the production line.

The acquisition and analysis of the image(s) can occur quickly (less than one second), which makes this technology applicable in industries where parts are produced on the order of seconds.

Machine vision has replaced human inspectors in many settings for obvious reasons. Automated inspection systems can run 24 hours per day, inspect complex parts quickly, and offer no bias as opposed to the human eye.

High-throughput screening (HTS) is traditionally recognized as a research activity, but in recent years, the development of more robust assays and instrumentation has largely industrialized this process. Pharmaceutical and biotechnology companies commonly have dedicated groups within their infrastructure that routinely perform HTS; groups that are actively seeking QC strategies to deliver shorter project completion times, more efficient use of reagents, and the generation of high-quality data.

One such quality-control strategy is the automatic calculation of data and its comparison to operator-defined acceptable levels. If the data falls outside of a predefined acceptable level, automatic notification to the operator is generated, typically via e-mail to a handheld wireless device.

HTS assays can vary in length from minutes to days. When assay times are longer, the automatic calculation of data as a process-control strategy may be useful but not sufficient to maintain an efficient process, especially since data collection occurs at the end of the process. For example, if an assay is five hours long and improper liquid additions occur within the first hour, that improper dispense could occur for up to four hours causing many units of production to be produced out of specification until the notification to the operator occurs.

The application of machine vision technology to HTS offers the ability to quality control this process in real time, allowing operators to respond to improper conditions in a timely manner conserving reagents, plates, and maintaining high-quality data.

Vision Inspection Systems

Scientists at Kalypsys and Merck developed a vision inspection system (VIS-1) (Figure 1) that can image a 1,536-well microtiter plate post dispense and evaluate if wells on the plate are empty or if there are droplets of fluid residing on top of the plate.

The system reports the number of empty wells and the number of droplets for that plate. The operator sets thresholds for these variables and will be notified in real time when a violation occurs. This strategy is effective since a poor dispense commonly leaves wells empty or fluid on top of the plate. The VIS-1 system is capable of imaging and analyzing all common plate types used in HTS.

Figure 1

VIS-1 utilizes two Cognex® cameras to image the plate. Post dispense, the dispenser software transfers control to VisiApp Software, which uses the cameras to take eight individual images, four per camera.

The cameras are mounted slightly offset from each other on a linear slide that moves over the top of the plate. Each image captures an area 16 by 12 wells in size. These eight images represent the entire plate. The cameras are built with an internal computer that holds the vision inspection program or job.

InSight Explorer was used to create the job. InSight is equipped with a powerful set of vision tools that are used to automatically align inspection regions along the bridges and well areas. Droplets, or blobs, are detected by inspecting the bridge region in a linear fashion for sets of connected pixels with a grey-scale value above (or below) a specified threshold.

Empty or partially filled wells are detected by analyzing histograms of the well regions. The histogram average and standard deviation serve as an indicator of well status (Figure 2). Threshold settings are determined prior to the HTS process by dispensing “good” and “bad” plates.

Figure 2

The bad plates should have empty and partial wells with significant droplets. The operator can generate bad plates by purposely misaligning dispense tips and unplugging the electrical lead to a solenoid valve so that no dispensing occurs producing an empty well. Typically, the empty well threshold is set to one, and the droplet threshold can vary depending on fluid dispensed and plate type. Total VIS-1 operation time is approximately 12 seconds.

Proof of Concept

In order to establish proof of concept, a cell-based calcium flux assay was performed. As the cells were dispensed, one of the tips was partially clogged (Figure 3). The VIS-1 detected a number of empty wells and defined a fail status. The bad wells had a smaller number of cells (microscope images) contributing to the lower cellular response (FLIPR data). This data is unacceptable and was detected by the vision system in real time. Had this been an actual HTS assay, the operator would have been notified and could have resolved this issue in real time.

Figure 3

VIS-1 is most effective on empty plate dispenses. If the plate already contains fluid, only droplet detection is effective. VIS-1 cannot currently detect differences in well volume. If a tip underdispenses without producing drops on top of the plate and the fluid covers the entire bottom of the well a fail status will not result. This presents an opportunity for those skilled in the art of machine vision to develop new image-analysis algorithms sensitive to detecting differences in volume height in a microtiter-plate well.

An alternative solution could be a gravimetric tip checking system used in parallel with VIS-1 to locate underdispensing tips. Gravimetric analysis of tip dispensing may require longer process times, which would decrease the frequency of this operation, i.e., every tenth assay plate the tips are checked for accuracy of volume dispensed.

VIS-1, a novel application of machine vision to the HTS process, can increase data quality, decrease wasted plates and reagents, and eliminate operator oversight of dispensing quality.

Jason Cassaday is senior research
engineer of automated biotechnology
at Merck & Co. Web: E-mail: [email protected].

Previous articleGenentech’s Option to Pursue Exelixis’ Oncology Candidate Triggers $3M Fee
Next articleConnecting Curious Cancer Clues