GFI | SCIENCE WATCH

Researchers used machine learning to screen 14 fungal proteomes for salty and saltiness-enhancing peptides. The discovery workflow is the real advance. Whether the peptides work in real foods, at a viable cost, remains untested.

Key Takeaways

• Jilin University researchers published an AI-guided peptide discovery workflow in npj Science of Food on October 20, 2026, screening peptides generated from 14 fungal proteomes.
• Six candidates were synthesized: three salty peptides and three saltiness enhancers.
• AHDQY gave the strongest salty response and ANSDL the strongest enhancement in electronic-tongue tests.
• The evidence is instrumental, cellular and computational, not consumer testing in real foods.
• The workflow may matter more than any single peptide, because it can shorten the search for taste-active ingredients.
• Enhancing saltiness does not replace salt’s technical functions, including preservation, water binding and protein functionality.

Why Sodium Reduction Is Still Hard

The World Health Organization estimates average global salt intake at about 10.8 grams per day, more than double its recommended limit of under 5 grams [2]. In many markets, processed foods supply most of it.

Salt is difficult to remove because it does more than taste salty. It lowers water activity, helps control microbial growth, solubilizes meat proteins, strengthens dough and steers fermentation. Existing tools each carry trade-offs: potassium chloride brings bitter or metallic notes, while yeast extracts and glutamates shift the flavour profile.

What the Study Did

The team built a dataset of taste-active peptides and compared three molecular embedding approaches, Uni-Mol, ESM-3 and MolCLR, within the same FT-Transformer classifier [1]. ESM-3, a protein language model, performed best and was used to screen peptides produced by simulated digestion of 14 fungal proteomes.

Six candidates were synthesized: the salty peptides AGHQR, AHDQY and SKPGNR, and the saltiness enhancers IPVVVDL, DNNEEW and ANSDL. In electronic-tongue analysis, AHDQY was the strongest salty peptide in deionized water, and ANSDL was the most effective enhancer in a low-salt sodium chloride system [1].

The peptides showed good compatibility with human endothelial cells and reduced high-salt-induced cell injury to varying degrees. Molecular docking and dynamics simulations suggested stable interactions with TMC4, a protein associated with salt taste perception [1]. The paper is an accelerated preview and may change before its final version.

The Method Matters More Than the Molecules

Conventional taste-peptide discovery is slow: hydrolyze a protein, fractionate, purify, then taste. AI screening reverses the order. Models rank thousands of candidates first, so only a handful need to be made and tested.

This study joins a fast-growing body of fungal taste research. Earlier in October, a separate npj Science of Food paper combined machine learning with sensory testing to identify peptides from the beech mushroom Hypsizygus marmoreus; the strongest raised perceived saltiness in salt solutions by about 71% [3][4]. Fungi are attractive sources because edible mushrooms, mycoprotein and fermentation biomass already sit inside food supply chains.

One caution: simulated digestion assumes these sequences would actually be released from real fungal proteins under industrial enzymatic conditions. That has to be demonstrated, not predicted.

What the Evidence Does Not Yet Show

Electronic tongues measure sensor responses, not human perception, and the study’s abstract reports no human sensory panel. Results in water and simple salt solutions may not carry over to foods, where fat, protein, acids and aromas can bind or mask taste compounds.

Commercial questions remain open. Peptides must survive heating, pH changes and storage without bitterness or off-notes. Cell compatibility is not a food safety assessment, and new taste-modulating ingredients would need regulatory review in most major markets. Chemically synthesized peptides are also far more expensive than salt, so a viable route would more likely be an enzymatic hydrolysate enriched in active sequences, which brings its own batch-to-batch variability.

Where It Could Fit

The most promising applications are soups, sauces, seasonings, savoury snacks and ready meals, where perceived saltiness drives acceptance and preservation can be managed through other hurdles. Cured meats, cheese and bread are harder, because salt there is structural and protective as well as flavourful.

GFI’s assessment is that peptides like these are best evaluated as part of a flavour system alongside potassium chloride and other enhancers, not as standalone salt replacers. As with any sodium reduction, shelf life and microbial safety must be validated independently.

AI will not take salt out of food on its own. What it can do is make the search for useful taste modulators faster and more targeted. The fungal peptides identified here are leads, not ingredients, and their next milestone is a consumer panel tasting real food.

Related on GFI: Rice Bran Could Help Foods Taste Saltier With Less Sodium. Can It Work Beyond the Lab? and Can AI Turn Food Factory Waste Into High Value Ingredients?

---

Sources

1. Zhang Y, Shen X, Liu Y, Li W, Han W. AI-guided discovery of fungal-derived salty and saltiness-enhancing peptides for sodium-reduced foods. npj Science of Food. 2026. Published October 20, 2026. doi:10.1038/s41538-026-01209-4. Journal page
2. World Health Organization. WHO Global Report on Sodium Intake Reduction. March 2023.
3. Kong W, Lin L, Lai L, Ding M, et al. Discovery of saltiness-enhancing peptides from Hypsizygus marmoreus via virtual screening, multisensory evaluation, electronic tongue, and molecular simulation. npj Science of Food. 2026. Published October 2, 2026. doi:10.1038/s41538-026-01088-9. Journal page
4. ScienceAlert. Adding 1 Mushroom-Derived Molecule to Salt Can Make It Taste 71% Saltier. October 2026. Article