Nearly all frozen bread dough products use chemicals like oxidizers, emulsifiers, and reducing agents to maintain strength after being frozen, thawed, and baked. The few frozen bread doughs that have been introduced with all natural ingredients have not been successful; consumers were not happy with product quality, and the doughs suffered from short frozen shelf life.
One company, Rich Products Corporation, set out to develop an all natural bread dough that would meet the same quality and taste standards as the company’s popular conventional bread doughs. Once it identified a number of possible natural ingredients, the company then faced the difficult task of optimizing the amount of each ingredient in order to meet specified requirements for flavor, appearance, and softness, among other qualities.
The company chose not to use conventional one-factor-at-a-time experimental methods because they would have taken an enormous amount of time. New optimal values would have had to be found for each variable every time another variable that interacted with it changed. Instead, Sachin Bhatia, a food scientist at Rich, used design of experiment (DOE) to simultaneously optimize the values of all factors. This statistical method helped him identify four different recipes that best met the company’s requirements. Bhatia tested each one before picking the best recipe for the final product. The result, which Rich believes is far superior to previous natural bread doughs, will soon be introduced to the market.
A Streamlined Design Process
Rich Products Corporation, a privately held international food products corporation headquartered in Buffalo, N.Y., faced a major challenge in extending its popular bread dough line with an all natural product. Bhatia felt that the problem lay with the development methods used to produce the dough. Because the amount of time required to make and test each recipe left only enough time to test a relatively small number out of the huge universe of possible formulations, he felt it might be possible to create a substantially better product by using a more scientific approach to fully explore the recipe space and optimize the formulation.
Bhatia decided to use Design-Expert software from Stat-Ease Inc., a developer of DOE software, to design an experiment to optimize the natural bread dough recipe. “I originally selected the software because it is designed for use by subject matter experts who are not necessarily experts in statistical methods,” he said. “The software walks users through the process of designing and running the experiment and evaluating the results.”
Mark Anderson, principal of Stat-Ease, says DOE drastically reduces the number of recipes that need to be tested by varying the values of all factors in parallel. “This approach determines not just the main effects of each factor but also the interactions between the factors,” he said. “As a result, DOE makes it possible to identify the optimal values for all factors with far fewer experimental runs than the traditional one-factor-at-a-time approach.”
Bhatia selected the D-optimal design because it provides the minimal number of blends ideally formulated to fit a given predictive model. He picked four different natural ingredients to include in the formulation of the new bread dough. The minimum value of each ingredient was set at zero to allow the experiment to explore the option of removing each ingredient from the recipe. Bhatia selected a quadratic model, which includes the non-linear blending terms for detection of component combinations that may be significantly antagonistic (detrimental) or synergistic (beneficial).
The factors and levels for the experiment were:
- ingredient A, 0 to 2;
- ingredient B, 0 to 4;
- ingredient C, 0 to 0.04; and
- ingredient D, 0 to 2.
The unit for each factor is flour percent or baker’s percent, which measures the weight of the ingredient relative to the total weight of the flour. The total of the level of the factors was fixed at four.
“One of the nice things offered by Stat-Ease is a consulting service in which they assist with designing the experiment and interpreting the results,” Bhatia said. “Since this was one of my first DOE projects, I decided to take advantage of this service. The design space looked unusual to me because the enzymes are used at very low concentrations, in the parts-per-million range. They looked over the design and confirmed that although the design space looked narrow, everything was correct.”
Bhatia asked the lab to produce all of the samples in a single day in order to reduce variability. Each sample was evaluated based on 10 different responses. Four of the responses were objective and included volume, length, the height of the bread after baking, and the width of the bread after baking.
A panel of taste testers subjectively evaluated the other responses. They considered overall likability, crust flavor, crumb flavor, softness, cell size, and crumb density. Bhatia entered the responses into Design-Expert, and the software generated a series of statistical analyses.
Bhatia used the analysis of variance results to verify the significance and accuracy of the model. The normal plot of residuals showed a high level of correlation between the data points, indicating that the results were internally consistent. Bhatia then developed a desirability function that provided weights and constraints for the responses, which he used to optimize the factors. The predicted response values for a most desirable formulation are shown in Figure 1 (above). It depicts settings and resulting predictions in the form of number lines superimposed on desirability scales, the higher the better. A 3-D rendering of the desirable region is shown in Figure 2 (left).
“I examined other desirable formulations to evaluate their distinctness and robustness as well as their overall likeability, which is the most important response,” Bhatia said. “I eliminated several because they were very close in the recipe space to other formulations that had higher overall likability. I also evaluated the robustness of the recommended formulations in order to account for manufacturing variability.”
Bhatia ordered samples of four selected recipes and did a follow-up sensory testing. Two of these samples scored highly; he then tested both of these formulations in the manufacturing process. The result? Based on manufacturability and cost, one of the four recipes was selected, and that recipe forms the basis of the company’s new line of bread doughs.