Optimization of fermentation conditions by Response Surface Method
In the process of constructing lycopene synthesis pathway, we found that its yield could not meet our expectations. Thus, we hoped to increase the yield from the point of fermentation. We chose to model the fermentation process and search for better fermentation conditions. We found the Response Surface Method (RSM) as a commonly used method for optimization.
Based on the experience of our previous fermentation experiments, we believe that the lycopene production is mainly influenced by the nutrient content of the medium, the oxygen content and the initial yeast amount. Thus, we chose to change the amount of supplied glucose, the level of dissolved oxygen, and the amount of inoculum to explore the optimal culture conditions.
We used the Box-Behnken Design to plan the experiments, and employed the Design-Expert 12 to analyze the results. We designed a three-factor, three-level experimental protocol according to the requirements. The specific factors and levels are shown in the following table.
We altered the concentrations of nutrients in the medium by supplementing different amounts of glucose into the YPD medium. We supplemented different YPD medium groups with 0, 4 and 8 g/L of glucose. Due to the limitations of available equipment, we were unable to quantitatively change the dissolved oxygen level. Nevertheless, we attempted to mimic diverse oxygen supplies by using different medium volumes in barbed conical flasks or plain conical flasks. Thus, with 25 and 50 mL of medium in two different barbed conical flasks, and 50 mL medium in a plain conical flask, we arbitrarily designated the dissolved oxygen levels in these three situations (or flasks) as 75, 50 and 25, respectively. The yeast concentrations at the beginning of fermentation varied by different inoculation ratios. We inoculated 2%, 4% and 6% of the yeast in proportion to different volumes of medium in each conical flask.
Based on the above factors and levels, the experimental protocol designed by the Design-Expert 12 is shown in the following table:
We used acetone to extract the lycopene produced by the fermentation of brewer's yeast. Then, the lycopene was determined using high performance liquid chromatography (HPLC). Each set of experiments was conducted in triplicates.
The results after the analyses are as follows.
We performed a multinomial regression on the experimental data to obtain the equation, as shown in Equation(1)
Y=1.195+0.089*A-0.009*B-0.04*C-0.003*A*B+0.015*A*C-0.0003*B*C-0.0061*A^2+0.0001*B^2+0.0046*C^2+0.00027*A^2*B-0.0024*A^2*C+0.00001*A*B^2 (1)The regression curve had a statistically significant p-value < 0.05.
The correlation coefficient of the regression equation was 0.9516, indicating that the linear relationship between the dependent variable and the three independent variables was significant and the model was a good fit.
Figure 1. The 3D diagram of the effects of dissolved oxygen and added glucose on lycopene production.
Figure 2. The 3D diagram of the effects of dissolved oxygen and inoculum size on lycopene production.
Figure 3. The 3D diagram of the effects of added glucose and inoculum size on lycopene production.
The image intuitively reflects the effects of the interactions among the three factors on the final yield. Comparing the three sets of graphs, we concluded that higher yield could be obtained when glucose was added in moderate amounts. However, this phenomenon was not evident when yield was generally high or low. The lower the dissolved oxygen level was, the greater its production could reach in general. The lower the inoculation rate was, the higher its yield could become.
Using the software to predict the optimal fermentation conditions, we determined that the optimal environmental conditions were 5.918 g/L glucose, 36.433 dissolved oxygen, and 2.122% inoculum to achieve the expected maximum yield. According to the actual situation, this experimental design could reach a yield of 1.3 mg/g DCW. This has indeed achieved a relatively high yield, proving the reliability of our model.
Optimal Conditions of Neural Network Model Prediction
After obtaining the data, we constructed a model based on Deep Learning. In short, this model used an artificial neural network, with the LeakyReLU as the activation function, the Mean Absolute Error as the loss function, the Adam as the optimizer, and the R2 as the metrics.
LeakyReLU is a commonly used activation function, and its specific formula is as follows:
LeakyReLU(x)=max(0,x)+negative_slope∗min(0,x)
The mean absolute error (MAE) is the average of the absolute deviation of all individual observations from the arithmetic mean. The average absolute error can avoid the problem that the errors offset each other, so it can accurately reflect the actual prediction error.
Adam is a first-order optimization algorithm that can replace the traditional random gradient descent process. It can iteratively update the weights of neural networks based on training data.
The following is the structure of the model:
After training model with the data, the model searches for the data existing in the interval with a fixed feature change, and a large number of predicted data can be obtained. Then, the 100 groups of data with the largest predicted value in the data are taken for visualization.
We found that our results were similar to those predicted by software
Visual recognition modeling
Filter out black areas based on RGB values of image pixels. Since the original shot does not appear completely pure black, so here, R, G, B values are taken as a smaller value to filter out areas close to black. After testing, the R, G, B values should be 50 when the screening effect is ideal.
Based on the analysis described above, only when the DNA assembly process is standardized, efficient and accurate, and the reaction conditions are simple enough, can the advantages of machinery be really implemented, and the efficiency of the Building module be improved. To achieve our goal, we employed the YeastFab technology from Tsinghua University, which was based on the Golden Gate DNA assembly method to enable the assembly of multiple DNA segments.
The shape of the petri dish in the shooting image is approximately circular. So, we used cv2.findContours() to extract the contour and used cv.drawContours() function to draw marks.
The culture dish lid is not allowed to be opened due to environmental restrictions during the test. The cv.circle ( ) function is used to reduce the range of the marked culture dish to remove the influence of the text on the color recognition of the colony. If there is no such restriction, the step can be skipped.
Load the yeast color template extracted on the original image in advance
Calculate the color mean of the yeast color.
Matching threshold: the test th takes 30 to separate two-color colonies better
Image staining: showing colony distribution.
Use cv2.findContours() to extract the contour
So far, the work of identifying, separating two color colonies from the original image and marking their center positions has been completed. Through the actual tests, the recognition accuracy is maintained at about 95%. If the difference between the two colors is more obvious (darker red), the accuracy can be improved.