Tofu is favored by people in East Asia and has been widely accepted by people from different areas of the world. Yield, texture, and protein content are major parameters used to evaluate the tofu quality. However, determining the parameters relies on tofu processing, which is time-consuming and labor-intensive. In addition, breeders/agronomists cannot get useful information from tofu parameters. Food scientists have related the tofu quality parameters to protein subunits of soybean seeds, such as 11S/7S ratio, 11SA3, and 11SA4 subunits. However, the profile of protein subunits has not been fully considered. North Dakota State University (NDSU) and Northern Crops Institute (NCI) plan to develop an alternative way to predict tofu quality based on the overall profile of soybean protein subunits. A high-resolution sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) will be employed to separate the protein subunit of 80 soybean seed varieties from East Asia and North America. Machine learning (ML) will be employed to construct a predictive model for the quality of tofu based on the SDS-PAGE images.

a. Classify soybean seeds based on their tofu quality, such as yield, texture, and protein content.
b. Test the protein subunit profile of soybean seeds
c. Build an ML model for predicting tofu quality based on soybean protein subunits
d. Evaluate the quality of soybean seeds from ND with the new ML model

1) The quality of tofu made from soybean from North America and Asia.
2) The different protein subunit profiles of soybean from North America and Asia
3) Build an ML model that can predict tofu quality based on the soybean protein subunits with an accuracy of > 95%.

Update:
The formal progress report has been uploaded as an attachment.

This project titled "A Novel Soybean Selection Method for Tofu Production Using Machine Learning" provides an overview of an ongoing research project conducted by North Dakota State University (NDSU) and Northern Crops Institute (NCI). The project aims to develop a new method for predicting tofu quality based on the overall profile of soybean protein subunits, using a high-resolution sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) to separate the protein subunit of 80 soybean seed varieties from East Asia and North America, and employing machine learning (ML) to construct a predictive model.

The objectives of the research are:
Classify soybean seeds based on tofu quality parameters such as yield, texture, and protein content.
Test the protein subunit profile of soybean seeds.
Build an ML model for predicting tofu quality based on soybean protein subunits.
Evaluate the quality of soybean seeds from North Dakota with the new ML model.

Key accomplishments to date include:
Collection and categorization of 178 soybean varieties from the United States (primarily North Dakota, Minnesota, and California) and China, spanning various latitudes and longitudes.
Hierarchical cluster analysis (HCA) of tofu prepared from these varieties, resulting in the categorization into six distinct clusters based on various parameters like water uptake, tofu yield, protein, firmness, and moisture content, among others.
Significant progress in SDS-PAGE, with the analysis of 80 varieties completed, identifying substantial differences in protein subunits between different soybean cultivars.
Development of a MATLAB-based algorithm to automatically read SDS-PAGE images, a preliminary step for the ML model.
Challenges faced include the unavailability of commercially pre-cast gels, requiring the team to make the gels themselves. The lack of skill among students in making these gels has led to some failures, slowing down the SDS-PAGE experiment progress compared to the tofu processing.

In summary, the study has provided valuable insights into how the source influences soybean seed characteristics and tofu quality. These findings have practical implications for the soybean and tofu industries, offering opportunities for product optimization and market differentiation based on sourcing. The project also emphasizes the importance of considering both soybean seed characteristics and tofu quality attributes in tofu production to meet consumer expectations and preferences.

View uploaded report

Updated June 24, 2024:

View uploaded report

This technology can benefit ND soybean growers by helping them understand the quality of their soybeans for tofu production and determine if soybeans intended for feed can also be sold at a higher price for the tofu process. This technology can also promote the screening of current ND soybean varieties that are suitable for making tofu and provide protein information for breeders to develop high-quality varieties for tofu products.