Korea innovation: AI model to speed up identification of new ingredients for health functional foods
Known as NaturaPredicta, the AI model could identify the functionalities and bioactivities of botanical ingredients without conducting preclinical or clinical tests.
It does so by comparing if an ingredient of interest could produce health benefits similar to ingredients that are currently approved for use in HFF in South Korea based on scientific evidence found on the PubMed database.
Using a scoring index, the higher the scores, the more similar the ingredients are in terms of their functionalities. Comparisons between the ingredients and their functions are also made using a heatmap.
The process uses AI techniques such as the natural language processing (NLP) and biomedical text mining tool BioBERT.
The concept is the brainchild of Dr Frank Kim, president of HFF, cosmetics and food regulatory consultancy firm SEAH Bio solution, which he has built using about two-and-a-half years.
The aim is to cut down the amount of time needed to assess the potential functionality of a botanical ingredient – a common difficulty faced by new ingredient developers all around the world.
“I was a research scientist in a local Korean pharmaceutical company where my role was in new drug discovery. I know how difficult it is and so, I need to support them from struggling. This is the purpose of building the model,” he said.
In its initial stage, NaturaPredicta studied the effects of nine ingredients against the existing 53 HFF ingredients in their functional benefits specifically related to chronic diseases.
The nine botanical ingredients are Arachis hypogaea, Avena sativa, Centella asiatica, Crataegus pinnatifida, Eleutherococcus senticosus, Helianthus annuus, Morus alba, Saccharum officinarum, and Triticum aestivum.
Findings from the AI model
Findings of the AI model were presented in the form of a heatmap, where darker hues signalled greater similarities between ingredients and vice-versa.
Based on the heatmap, peanuts (Arachis hypogaea), was the most similar to two existing HFF ingredients, namely ginger (zingiber officinale) and turmeric (curcuma longa), with a score of 0.953 and 0.946.
This indicates that peanuts contain bioactives that have similar functions as ginger and turmeric.
Similar scores were also seen in their functions related to blood glucose, blood triglycerides, and body fat.
In addition, peanuts and turmeric showed similarities in their functions related to blood glucose, cognitive ability, and female menopause.
“In this study, NaturaPredicta™ demonstrated its possibility to be used for finding similarities between HIs (Health functional food Ingredients) and potential HFF candidates intended to be developed.
“Our result can suggest its applicability in providing guidelines for identifying relevant bioactivities,” said Dr Kim and his fellow researcher Daeyong Choi in the findings published in Food Supplements and Biomaterials for Health.
NaturaPredicta is currently patent pending in South Korea.
Interest from the pharmaceutical firms
The model is seeing interest from pharmaceutical firms that also operate a health functional foods division, Dr Kim noticed, following his presentation on the model at Vitafoods Asia in Bangkok last month.
A reason is because the AI technique is already widely used in new drug discovery and so, pharmaceutical companies are familiar with the concept as opposed to food companies.
However, the process of new ingredient discovery for pharmaceutical drugs and health supplements are entirely different, he cautioned.
This is because active pharmaceutical ingredient (API) involves single molecule, while health functional food ingredients are more complex in that there are tens and thousands of active compounds in an extract.
Take green tea extract as an example, he said that while catechin was a well-known polyphenolic compound found in green tea, there were several other compounds present in green tea that work together to produce the health benefits.
In other words, the health benefits of the green tea extract are not based on catechin alone.
“Catechin is just one compound of the green tea. That’s why sometimes there are studies which showed that catechin supplementation alone did not produce health effects similar as green tea.
“This is because there are so many huge numbers of molecules that associate and interact between one another and they come to the body and nobody knows exactly how they interact to produce the benefits,” he said.
“Green tea is green tea, green tea is not catechin,” he added.
Helping the farmers
Aside from new ingredient developers, the AI model could also help Korean farmers who are keen to find out the potential of their crops as functional food ingredients.
However, their biggest frustration is the lack of capabilities in discovering and testing the potential benefits of their crops.
Often times, these farmers would pay researchers and scientists to find out the potential health functions of their crops without any guaranteed results.
“The farmers want to develop their crops as functional food ingredients, but many are not scientists, and they will just go to the professors and pay them a large amount of money to do the testing.
“In the end, they could be told that there’s no statistically significant results shown, but they will still need to pay the balance for the research.
“This is really happening in South Korea and our goal is to reduce the unnecessary expenses and time for these farmers,” Dr Kim said.
What’s next?
The next step, according to Dr Kim, is to apply the AI model in the study of more ingredients for use in HFF.
These ingredients would include those permitted by various regulators across the world, including the Korean FDA, the US FDA, and those registered with European Food Safety Authority (EFSA).
“We can get at least over 9,000 of the ingredients, so that we can have a bigger database and add more insights for the developers of new health functional food ingredients,” he said.
In South Korea, the authorities have listed out 31 types of functional claims that companies could apply to HFF.
The functionalities cover seven body systems, including the cardio-circulatory, the digestive, the urogenital, and the bone-muscle system.
For example, functional claims under the cardio-circulatory system could involve the blood circulation, blood pressure control, and cholesterol control.
A total of 53 HFF ingredients have been identified by the Korean authorities in their ability to produce the 31 types of health benefits.
The 53 HFF ingredients comprised of 21 notified HFF ingredients and 32 HFF ingredients that are individually recognised by the government.
