如何高效提升原物料建模效率、降低實驗成本？定量構效關係 (QSAR) 性能優化 4 步驟

[#1]

In the field of material research and development and manufacturing, experimental design is a key step in evaluating raw material properties and process conditions. However, traditional experimental methods will significantly increase the number of experiments as the number of raw material types increases, leading to an increase in time and cost. In addition, when new materials are added, the experimental design often needs to be redone, increasing the uncertainty of R&D. To solve this problem, the QSAR (Quantitative Structure Activity Relationship) method can be introduced into the experimental process to improve the efficiency of raw material screening and analysis through data modeling, thereby reducing experimental costs and improving decision-making accuracy.

What isQSARmethod?QSAR Application of the method in raw material screening

QSAR (Quantitative Structure-Activity Relationship) is a method to establish a mathematical model based on the relationship between the molecular structure and activity of a compound. This concept was first proposed in the 18th century and is now widely used in drug development, chemistry and materials science. Traditional QSAR is mainly used for the study of the relationship between chemical structure and biological activity, but this case expands its application scope to physical and chemical properties, using molecular structure data to improve the efficiency of raw material screening and modeling.

Case: From 45 Raw materials selection 12 Experimental design

A company wants to conduct formulation experiments on different raw materials, but due to cost considerations, it needs to reduce 45 raw materials to 12 . To this end, the team used the QSAR method to screen through 19 molecular structure characteristics and 2 key Y variables and designed the best experimental plan. The following are the specific experimental steps:

Step 1 : Convert categorical factors to continuous factors

Since the chemical properties of raw materials are categorical variables, they need to be converted into quantifiable continuous variables through professional knowledge and data analysis experience, as shown in ( Figure 1) . For example, based on molecular structural characteristics, key predictive variables are extracted to enable subsequent modeling.

(Figure 1: Each material has corresponding molecular structure data )

Step 2 : Dimensionality reduction and decorrelation using principal component analysis (PCA)

In multivariate analysis, many variables may be highly correlated ( Figure 2) , affecting the accuracy of modeling. Therefore, the research team used principal component analysis (PCA) to transform the variables into independent variables, as shown in ( Figure 3) , to reduce the dimension and avoid collinearity problems. This process helps in subsequent modeling and variable selection.

(Figure 2) Correlation analysis

(Figure 3) Principal component analysis

Step 3 : Design the experiment through Custom Design

In JMP , the Custom Design function ( as shown in Figure 4) was used to design the optimal experimental plan using the principal components transformed by PCA as covariates. In this way, the research team successfully controlled the number of experiments to within 12 times, significantly reducing costs and time requirements.

(Figure 4-1)

(Figure 4-2)

(Figure 4-3)

Step 4 : Apply PLS or machine learning for modeling

After completing the experimental design, use the PLS method to select a 6- factor model based on Prob > van der Voet T ² ( as shown in Figure 5) , or use the Model Screening platform ( as shown in Figure 6) to screen out suitable predictive modeling methods, save the formula to the data table, and use the Profiler under Graph ( as shown in Figure 7) to understand the relationship between the X variable and the Y variable.

Figure 5 PLS platform

Figure 6 Model screening platform

圖7 Profiler

summary

This case proves that applying data modeling technology to raw material screening can significantly improve R&D efficiency and provide more possibilities for future process optimization.

Through this case, researchers can expand variables to conduct experimental design together with process variables. In addition, using the QSAR method when screening material types not only reduces the number of experiments but also enables reverse search for materials, which can significantly improve the efficiency of raw material screening and avoid waste of experimental costs, providing more possibilities for future process optimization. With the continuous advancement of materials science and data analysis technology, this method will become an important tool for companies to improve their R&D decision-making capabilities and the key to future industrial upgrading.

References

1. https://community.jmp.com/t5/Discovery-Summit-Europe-2017/Increase-Efficiency-and-Model-Applicability-Domain-When-Testing/ta-p/36572 2017 Presentation
2. https://community.jmp.com/t5/Learn-JMP-Events/JMP-Academic-Webinar-Teaching-Analytics-in-Chemistry-and/ev-p/481604 JMP Academic Webinar - Teaching Analytics in Chemistry and Chemical Engineering with JMP: A Hands-On Introduction Case2

More learning resources

• Application of machine learning in chemical industry and manufacturing industry
• Application of JMP in the chemical industry: two case studies
• DOE classic case practice: 3 experimental design methods for product development and process optimization
• Download JMP for free and try it for 30 days