🛠️ This page is under construction 🛠️

Find and use connections between nutrients

The Cedar Nexus project is the synthesis of two initiatives I am currently working on. It's a long-term effort to understand how nutrients interact with each other and how this knowledge can be applied to create a more sustainable and healthy food system. The core idea is to identify the most important connections between nutrients and to develop a method for predicting and optimizing these interactions. The project is still in its early stages, but I am excited to announce that I have made significant progress in this direction by training a machine learning model to predict unknown nutrient contents.

layers

Nutrideep

Nutrient mapping using machine learning prediction

  • Barcode input of the food label
  • automation prediction of missing nutrients
  • output_circle output of a full nutritional profile

Progress

90%
  • find a suitable regression logic
  • write an algorithm
  • train models for amino acids
  • train models for minerals
  • train remaining models
  • write an api to access the models
read more
graph_3

Nutrimerge

Inclusion of emergent processes and dynamic interactions of nutrients

  • calculate dynamic result-driven computations
  • insights insights from multiple studies
  • output_circle output of accurate nutrition values

Progress

65%
  • invent a serial bus system
  • collect scientific literature
  • reconstruct functions from results
  • write algorithms
  • repeat previous steps
  • connect with Nutrient Mapping
  • create a UI
read more

Nutrideep - A Nutrient Prediction Tool

Predict Missing Nutrients

By leveraging deep learning techniques, I have developed a tool that predicts missing nutrients from a given set of known nutrients. For example, if you know the energy, protein, fat, and sugar content of a food item, the model can predict the levels of missing nutrients such as vitamins, minerals, and amino acids with remarkable accuracy.

The predictive model analyzes known data to fill in the gaps in nutritional information, providing a much more comprehensive profile of a food item. This helps users understand not just what they eat, but how it contributes to their overall nutritional needs.

How it works

  • Input: Nutrient data (e.g., calories, protein, fat, sugar, salt) from food packaging.
  • Prediction: The machine learning model predicts missing nutrient values, such as specific vitamins or amino acids.
  • Output: A fuller, more accurate nutritional profile of the food item, including predicted values.

The model's high accuracy (with an R² of 0.97 for predicting amino acid content) makes it a valuable tool for anyone seeking a more complete understanding of their nutritional intake.

How it performs

The following diagram shows the predicted values tested against actual values for the amino acid serine. The high accuracy of the model can be seen by the clustering of data points around the 1:1 line, indicating that the model is able to accurately predict the values of serine up to 20g per 100g of food.

Predicted values tested against actual values for serine

The model's ability to predict values across a wide range of concentrations is important, as it means that it can be used to accurately predict the nutrient content of a wide variety of foods.

As with any machine learning model, there is always room for improvement. There are still some outliers that the model struggles to predict accurately. These discrepancies highlight areas for further refinement and optimization of the model. By continuously updating and validating the model with new data and refining its predictions, its performance can be enhanced, ensuring more reliable predictions across diverse datasets.

In addition to amino acids, the model also predicts the nutritional values of vitamins and minerals, which can be useful for food consumers and health professionals.

Use Cases & Applications

The ability to predict missing nutrients has numerous practical applications, which can benefit different groups of people in various fields.

person

Consumers & Health Enthusiasts

Consumers can use this tool to gain a better understanding of their diet and ensure they're meeting their nutritional needs more accurately.

fitness_center

Athletes & Bodybuilders

Athletes can benefit from knowing their full amino acid and micronutrient intake to optimize performance, muscle growth, and recovery.

local_dining

Dietitians & Nutritionists

Health professionals can enhance their dietary recommendations by accessing more comprehensive nutritional profiles for their clients.

storefront

Food Manufacturers

Food companies can improve product formulations and nutritional labeling by using predictive models to enhance nutritional transparency.

apps

Health & Wellness Apps

Apps can integrate this tool to provide users with detailed, predictive nutritional insights, allowing them to track more comprehensive food data.

Delve deeper into the network of nutrient interactions

The predictive model is just a starting point. Understanding the relationships between nutrients may contribute to improvements in food development, health, and sustainability.

Exploring nutrient interactions could help us realize the potential of nutrient prediction, possibly aiding advancements in these areas.

Nutrimerge - A Network of Nutrient Interactions

Focus on dynamic interactions

Just as a tree is much more than its roots, branches and leaves, the nutrients you digest are not just the sum of what you eat. Dietary fiber, proteins, carbohydrates, fats and various micronutrients interact with each other and impact absorption to a varying degree.

Traditional approaches simply add together the ingredients of the food consumed. In terms of calories, that would mean 575 Kcal in a simple example of an avocado and some broccoli.

Avocado and broccoli = 575 Kcal ❌

However, this is not what happens in real life, where emergent processes from dynamic interactions of the food matrix play an important role. My approach tries to cover as many known interactions as possible in this way:

Avocado and broccoli Nutrients such as fiber, calcium, protein and fat dynamically inhibit each other = 430 Kcal ✔️

Just looking at the interactions between the four nutrients - fibre, fat, protein and calcium - makes a significant difference and results in 430 instead of 575 Kcal in this example.

Observing nutritional emergence
through a scientific lens:
The maths

To begin with, I have designed a serial bus system, where the calculations for each interaction are carried out one after the other. Each calculation is performed using the results of the previous calculation.
These calculations are based on the averages of many study results, reconstructed into mathematical functions. For very complex interactions, piecewise functions are used.

The following diagram tentatively shows how the intake of total energy, protein and fat is affected by the amount of fibre consumed, based on the results of various studies (Wisker et al., 1988, Baer et al., 1997, Corbin et al., 2023).

Nutrients such as fiber, calcium, protein and fat dynamically inhibit each other

Another illustrative example is the phenomenon that calcium undergoes a saponification reaction with fat in the intestine, the result being a poorly soluble reaction product that is excreted undigested. However, even normal amounts of ingested protein break down this structure. The effect is shown in this diagram, which takes its data from an example meal:

Nutrients such as fiber, calcium, protein and fat dynamically inhibit each other

The underlying calculations, reconstructed from several study results, are as follows:

X = 14.18 + − 8.57 1 + ( Ca 1519.93 ) 2.39
  • X = Fat excretion (g)
  • Ca = Daily calcium intake (mg)
Protein factor = 0.0074269 + 0.99256 1 + ( P energy% 20.760 ) 22.20
  • Penergy% = Percentage of protein in total energy
(Christensen et al., 2009)

Contact

If you're interested in collaborating, contributing to the project, or simply learning more, feel free to get in touch! Let's make nutrition science more accessible and impactful together.

emailcedarnexus@gmail.com

Get in touch

Write me a message:

Name
Email
Message

Cited Sources

Christensen, R., Lorenzen, J. K., Svith, C. R., Bartels, E. M., Melanson, E. L., Saris, W. H., Tremblay, A., & Astrup, A. (2009). Effect of calcium from dairy and dietary supplements on faecal fat excretion: A meta-analysis of randomized controlled trials. Obesity Reviews, 10(4), 475–486. https://doi.org/10.1111/j.1467-789X.2009.00599.x
Wisker, E., Maltz, A., & Feldheim, W. (1988). Metabolizable energy of diets low or high in dietary fiber from cereals when eaten by humans. The Journal of Nutrition, 118(8), 945–952. https://doi.org/10.1093/jn/118.8.945
Baer, D. J., Rumpler, W. V., Miles, C. W., & Fahey, G. C. (1997). Dietary fiber decreases the metabolizable energy content and nutrient digestibility of mixed diets fed to humans. The Journal of Nutrition, 127(4), 579–586. https://doi.org/10.1093/jn/127.4.579
Corbin, K. D., Carnero, E. A., Dirks, B., Igudesman, D., Yi, F., Marcus, A., Davis, T. L., Pratley, R. E., Rittmann, B. E., Krajmalnik-Brown, R., & Smith, S. R. (2023). Host-diet-gut microbiome interactions influence human energy balance: A randomized clinical trial. Nature Communications, 14(1), 3161. https://doi.org/10.1038/s41467-023-38778-x