Genetic Engineering News. Volume 18, Number 20, November 15, 1998
By Vicki Glaser

Trapped within the cyberspace (or paper fanfolds) of bioprocess databases is a wealth of data that define the optimum settings of critical parameters for a manufacturing process. The analysis of individual parameters, or of the performance of a single batch, yields information that can contribute to overall process improvement.

Aegis Analytical Corp. (Eastlake, CO) has designed Discoverant, a software system that helps extract information from disparate sources of manufacturing data in an interactive environment, and utilizes pattern recognition, multivariate analysis and other statistical tools to correlate key process parameters to outcomes. Aegis’ vice president of applications, Justin 0. Neway, Ph.D., defines the field of manufacturing informatics as "a way of looking at raw manufacturing data and turning it into information."

Discoverant provides access to all the different data associated with manufacturing runs, allowing for a retrospective analysis to identify the correlations between combinations of process variables and desired outcomes. The software also facilitates the process of troubleshooting a manufacturing process that drifts out of control or starts to fail final product specifications. Aegis is developing Discoverant in partnership with Genentech (S. San Francisco), Merck (West Point, PA) and Hoechst Marion Roussel (Frankfurt, Germany).

Dr. Neway describes a "disconnect" between the control variables and settings used by manufacturing process control systems "and the closure of the loop, where you learn something about how the process actually performed, and then go back and put in better control settings to achieve better yield or higher quality product."

"It is the interrelationships between the different kinds of data that are interesting, more than just the raw data itself," says Peter Salmon, director of fermentation process development at Merck Research Labs (Rahway, NJ). It would take weeks to collect all of that data from the various data management systems operating in a manufacturing facility: the Laboratory Information Management System (LIMS), online and offline process data, raw materials data, maintenance data on the instruments and equipment, culture stock and operators on duty are just some examples of the varied data sources. Salmon describes these as "islands of highly automated data collection," each managed by independent computer systems. By bringing all of this data together, you can" interpret the common themes that are behind good batches and bad batches," he says. You can translate this data into information.

"In the 1980s, we moved from manual data collection to automated data collection," explains Salmon. The next step is to move to "automated management and interpretation of that data," looking for correlations. Abandoning tried and tested information or control systems, and having an all-encompassing computer system, is not the only solution. The Aegis system has "a middle layer that will pull the data from all these diverse sources and present a uniform interface."

Conformance and Optimization

In bioprocess applications, manufacturing informatics has a role to play not only in the development of a new process, but also in the optimization of an approved, commercial - "locked in" - process. In the eyes of the FDA, "the process is the product" explains Dr. Neway. "Not only do you need to meet final specifications, but you have to show that your process conformed with your own constraints with respect to the process parameters." For each batch, the critical parameters must stay within their defined ranges. Data downloaded from the process controllers, supervisory control and data acquisition (SCADA) systems or distributed control systems (DCS) - either paper printouts or electronic batch reports - must be scrutinized to check the conformance of the critical process parameters.

The electronic data needed to demonstrate process parameter conformance is present in various databases throughout the process system; however, few nonprogrammers are capable of writing the necessary queries in command line code to access this information. One advantage of the Discoverant system is its point-and-click browser interface and built-in data facilitator. This gives all members of the operations and troubleshooting team, from process engineers to biologists, chemists and technicians, access to the process data in place, and minimizes manual data management tasks. Discoverant's data input capabilities can also ease the ongoing transition from paper to electronic batch records in the bioprocess industry.

Some degree of variability exists even in mature, commercial processes, as the critical parameters fluctuate within their approved ranges. This variability will affect outcomes, in particular, product quality and yield, which have economic implications that affect the manufacturer's ability to compete in the marketplace. The various allowed combinations of parameter settings used throughout a batch run will result in a distribution of yield and quality that can vary by as much as 10% - 20% during a given manufacturing campaign.

Dr. Neway suggests that if you "tweak" the process, creating a better overall combination of process parameter settings, while keeping each within its allowed range, you can bias the operation to achieve top performance in yield and quality for a higher proportion of the time. Defining the optimal combination of variables and settings, though, is a daunting challenge. Whereas establishing an association in a fermentation operation between a single process parameter and growth rate or yield is relatively straightforward, determining how that parameter relates to the combined product recovery and downstream purification processes is far more complex.

The goal is to improve product yield or quality in a commercial process without changing the manufacturing technology itself. Once a correlation is identified and quantified, the user can adjust particular parameters to maximize yield and quality in future batches. Because the parameters do not exceed their approved ranges, the manufacturer would most likely not have to perform equivalence studies or repeat clinical trials to meet FDA regulations, provided the product passes specifications.

Embedded in the Discoverant system is a powerful statistical package that provides the tools and workflows for performing multivariate analysis, principal component analysis and other pattern recognition techniques. With these tools, Discoverant can determine "what combination of process parameter settings, across all unit operations, gives the best outcomes," says Dr. Neway. Aegis embedded the statistics engine behind an Internet browser interface that facilitates data access, analytical operations and reporting.

In March 1998, Aegis announced an agreement with Visual Numerics, Inc. (Houston, TX) to integrate that company's JWAVE Internet-enabled data visualization tool with Discoverant. JWAVE is a set of Java components used to build visual and numerical applications in Java, including the creation of 2D, 3D and 4D plots, animations and images. In Discoverant, these tools are present in ready-to-use applets for retrospective data analysis tasks.

Animated Process Signatures

Discoverant's advanced visualization capabilities allow the user to display bioprocess data in 3D and 4D images. The software can be used to create a "process signature," an electronic batch report that is a graphical and numerical representation of all the critical process parameters for a particular product batch or group of batches. By displaying the parameters multidimensionally, Discoverant presents a "high information content snapshot" of a batch or group of batches, allowing the user to make statistical comparisons between batches and manipulate the data.

By comparing data from multiple batches, the user can identify repeating patterns between runs. Such patterns suggest that "there are practices somewhere in the process that have periodicity," says Dr. Neway. If those periodic practices correlate to process outcomes, then the user would want to identify and "stabilize" them to favor the desired outcome. Discoverant employs pattern recognition and multivariate analytical tools to establish these correlations and to quantify the relationships between process parameters and outcomes.

Discoverant incorporates both continuous data from SCADA systems and discrete point data from manufacturing databases and paper records into the process signature. The software can create multipanel displays, combining critical trend data in formats such as scattergrams, histograms, line plots, cusums and event charts. The process signature can serve as a tool for minimizing batch-to-batch variability, demonstrating in-process parameter conformance improving and troubleshooting a manufacturing process (to prevent batch failure) and establishing a new process through comparative analysis of process signatures to facilitate technology transfer.

Aegis can take this multidimensional data representation a step further and create an animated image of a rolling average of several batches over the course of a manufacturing campaign. Each frame in the animation is the next group of batches in the series. "You can watch it as a movie," explains Dr. Neway. "You can watch, over process history, what's changing in the batch process signature, the relative standard deviations [representing variability in the process parameters], the means and the principal components that are most affecting outcomes." This historical account allows the user to see the effects on the process of factors that vary over time, such as seasonal changes, which can introduce variation in the mineral content of the water or in the humidity of the air, for example. Another useful application of this multidimensional image animation technique is to observe, simultaneously, the changes in all the process parameters as a progression from the worst-performing batches to the best-performing batches. This lets the user literally "see" the key differences between bad and good runs.

Founded in 1995 by Gretchen L. Jahn, Co-Founder, Chairman of the Board, Aegis Analytical is privately held, relying on private sources of funding while completing its first round of venture capital financing. The company has a network of 20 people, both full-time employees and personnel doing contract research for Aegis through established corporate alliances and strategic partnerships. Approximately 12 new employees will join the company once the financing is completed.

Aegis designed Discoverant as a stand-alone manufacturing informatics system for retrospective batch data analysis to guide process improvement and troubleshooting, and for demonstrating in-process parameter conformance.