Quantifying Nutrient Profile Similarity Among Foods Using Cosine Similarity of Multidimensional Nutrient Profiles

Background: Food substitutions can be used for improving dietary performance in different domains, e.g., human health, environmental impact, affordability or cultural preferences. However, the criteria to justify food substitutions are often inconsistent and based on single nutrients such as maintaining volume, protein, energy or culinary equivalence. Such approaches neglect the broader and more complex nutrient profiles that define a food’s overall nutritional value. Methods: This study introduced the application of cosine similarity to enable comparisons of nutrient profiles of foods based on a multitude of nutrients as an important step in food substitutions. Each nutrient was normalized to a fraction of the Reference Daily Intake and represented an axis in a multidimensional space. Subsequently, each food in the dataset (n = 927) was turned into a vector in this space, and the cosine similarity between each pair of vectors was calculated. The effect of nutrient selection was assessed by using six nutrient subsets, differing from 6 to 21 essential nutrients. Results: Cosine similarity provided a quantitative measure of nutrient-profile similarity and revealed similarities across food pairs. For most food categories, we found a broad distribution of cosine similarity values, spanning from 0 to 1. The choice of nutrient subset shifted the results, highlighting the relevance of including a wide range of essential nutrients. Conclusions: We conclude that cosine similarity is a valuable tool in identifying similar nutrient profiles, offering a comprehensive and flexible framework, complementary to nutrient density approaches, for guiding dietary choices across multiple objectives without compromising nutritional quality.