In a meringue, protein helps hold a structure full of air. The amount on a nutrition label cannot explain how well it will perform that job. To choose a replacement, a food developer also needs to understand its behaviour inside the mixture.[3]
AlterProtX, developed by Eunyeong Lee and Hee Yang at Kookmin University in Seoul, approaches that problem through molecular comparisons. Their 8 October study screened about 1.35 million proteins from fifty non-animal sources against animal-food references.[1]
An egg contains a crowd of proteins
The target is already complicated. A separate 2023 proteomics study analysed forty-five eggs and identified 398 proteins in the white and 456 in the yolk. Its work illustrates how much variety sits behind the everyday phrase egg protein. The two compartments contain different molecular collections as well as different proportions of nutrients.[4]
Heat can change a protein's shape and whether it stays dissolved. In the Meltome Atlas research underlying the thermal training data, scientists heated cells or cell extracts and measured soluble protein. The endpoint reflects unfolding and precipitation together, rather than simply the melting point of a purified ingredient.[5][2]
AlterProtX combines sequence features with a graph of neighbouring amino acids to predict thermal stability, then compares distributions of thermal and interaction properties across sources. Its rankings prioritise experiments; they are not demonstrated replacements in cooked food.[1]
The thermal model used a curated dataset of 24,572 proteins. Its reported test-set mean absolute error was 4.14°C, compared with 4.24°C for the complete-sequence model without a structural graph, in fivefold evaluation. This measures temperature-prediction error, not performance in a recipe.[2]
The distributional approach matters because an average can conceal a mixture. As an illustrative example, a source containing both heat-sensitive and heat-resistant proteins could have the same mean thermal stability as one whose proteins cluster in the middle. The ingredients would present different populations of molecules as the temperature rose.
Different targets, different matches
Three leading matches among the fifty sources
| Animal reference | Highest-ranked source | Additional finding |
|---|---|---|
| Egg white | Wheat | Low lysine relative to the cited amino-acid reference; established allergenicity.Source 2 |
| Egg yolk | Quinoa | Essential-amino-acid profile met or exceeded all nine categories of the cited reference pattern.Source 2 |
| Meat | Cauliflower mushroom (Sparassis crispa) | Comprehensive amino-acid composition data unavailable.Source 2 |
Wheat’s proteins resembled egg white in heat response, molecular mass, surface affinity for water and the pH at which they carry no net charge. Those features affect interactions within a mixture. Quinoa came closest to yolk’s combined profile. Changing the target changes which source comes first.[1]
The table exposes a useful separation between function and nutrition. A candidate can rank well for the physical behaviour a developer wants and still leave an amino-acid gap to address. Conversely, a nutritionally appealing source may require more work to provide the right structure. There need not be one winner for every purpose.
AlterProtX weights each protein equally because consistent abundance data are unavailable. Tissue, extraction and processing affect which proteins dominate an ingredient, so this source-level comparison precedes formulation.[1]
What happened when researchers baked the wheat
In 2018, Arno Wouters and colleagues at KU Leuven incorporated enzyme-treated wheat gluten into a meringue recipe. Their published report describes lower-density batters and greater baked volume per unit mass than the egg-white-only comparison, without notable differences in colour or texture.[3]
The preparation mattered: the researchers used hydrolysed gluten, with protein chains cut by enzymes to improve solubility and foaming. Egg-white solutions held foam more stably without sugar, but that advantage mattered less in the viscous, sugar-rich batter. The whole mixture changed the importance of an isolated property.[3]
The next batch of meringues could test choices the ranking leaves open: which wheat proteins are extracted, how extensively enzymes cut their chains and how much sugar goes into the mix. For food developers, the shortlist becomes valuable when it leads to those precise formulation decisions.
Sources & further reading
- An AI-enabled proteome-scale framework for identifying sustainable protein alternatives for future food systems npj Science of Food · 8 Oct 2026 · Accessed 18 Oct 2026
- Supplementary Information: An AI-enabled proteome-scale framework for identifying sustainable protein alternatives for future food systems npj Science of Food · 8 Oct 2026 · Accessed 18 Oct 2026
- Enzymatically Hydrolyzed Wheat Gluten as a Foaming Agent in Food: Incorporation in a Meringue Recipe as a Proof-of-Concept Journal of Food Science · 23 Jul 2018 · Accessed 18 Oct 2026
- Egg White and Yolk Protein Atlas: New Protein Insights of a Global Landmark Food Foods / PubMed Central · 18 Sep 2023 · Accessed 18 Oct 2026
- Meltome atlas—thermal proteome stability across the tree of life Nature Methods / Utrecht University Repository · 13 Apr 2020 · Accessed 18 Oct 2026