Machine learning and molecular simulations reveal novel umami peptides in Dengchuan beef

Scientists in China have decoded the molecular secret behind the famously savory taste of Dengchuan beef, a prized local cattle breed from Yunnan Province, and in doing so they have uncovered a family of long-chain umami peptides that taste more potently than monosodium glutamate itself. The research, published in Current Research in Food Science, combines peptidomics, machine learning, sensory science and molecular simulation into a single pipeline that whisked thousands of candidate molecules down to just five standout flavor compounds. The work offers one of the most detailed looks yet at how long-chain peptides, rather than the short ones that dominate the literature, activate the human umami receptor.

Umami, often described as the fifth basic taste alongside sweet, sour, salty and bitter, is the savory depth that makes broths, cured hams and slow-cooked meats so satisfying. The main umami substances in food include free amino acids, organic acids, nucleotides and umami peptides. Over 200 umami peptides have previously been extracted from proteins in chicken, fish and goose, but most research has focused on short chains of fewer than ten amino acids. The taste behavior of longer peptides, which may persist on the palate longer and bind to receptors in more complex ways, has remained largely unexplored. Beef was an obvious place to look: the umami precursor inosinic acid was first isolated from beef soup, and Yunnan’s local cattle are renowned for tender meat and delicious broth, yet the peptide contributors to Dengchuan beef’s characteristic flavor were unknown.

The team began by preparing a water extract from fresh Dengchuan beef, homogenizing the meat with water at a one-to-three ratio, heating it to 90 degrees Celsius for 30 minutes, then cooling, centrifuging and filtering the supernatant. Ultrafiltration membranes with cut-offs of 3 and 5 kilodaltons split the extract into three fractions. A trained ten-member sensory panel, all food science professionals at Yunnan Agricultural University, scored each fraction for the five basic tastes. The smallest fraction, containing peptides under 3 kilodaltons, scored dramatically highest for umami at 7.09 out of 10, significantly outpacing the 3-to-5 kilodalton fraction at 5.36 and the largest fraction at 3.64. That result aligned with earlier findings that low molecular weight peptides are the strongest umami carriers, so the smallest fraction advanced to structural identification.

Using nano-flow liquid chromatography coupled to a Q Exactive HF-X mass spectrometer, and processing the raw data with MaxQuant software at a false discovery rate below one percent, the researchers identified a remarkable 4,220 peptides. Molecular weights ranged from 349 to over 3,100 daltons, with nearly 44 percent of the peptides falling below 1.5 kilodaltons, a size class previously associated with umami activity. Most of these smaller peptides contained 9 to 15 amino acid residues, making them notably longer than the dipeptides and tripeptides that typically dominate umami studies. Sequence analysis revealed telling patterns: alanine, aspartic acid, glutamic acid, leucine, serine, threonine, valine and glycine crowded the N-termini, while arginine, lysine, asparagine and leucine anchored the C-termini. Sixty-five peptides combined an N-terminal glutamic acid with a C-terminal lysine, leucine, threonine or arginine, a physicochemical signature known to boost umami. Ultimately, 143 peptides in which umami-related acidic amino acids, aspartate and glutamate, made up at least 30 percent of the sequence were selected for the next round.

Rather than relying on any single prediction tool, the team deployed a three-model machine learning gauntlet: iUmami_SCM, Umami_YYDS and Tastepeptides_DM, requiring an iUmami_SCM score above 588, umami probabilities above 0.9 and bitterness predictions below 0.1. This cross-validated strategy, previously used to screen 155 candidate peptides from button mushrooms, trimmed the pool to 98 peptides and then, with a stricter iUmami_SCM cutoff above 650, to 23 peptides mostly 9 to 13 residues long. Safety and solubility filters followed. One candidate, AEEEYPDLSKHN, was flagged as toxic by the ToxinPred platform and discarded; the remaining 22 were all predicted non-toxic with good water solubility, and BIOPEP analysis suggested umami-active fragments made up more than half of each sequence.

Molecular docking then served as the final computational sieve. The researchers built a homology model of the human T1R1/T1R3 taste receptor, the principal umami detector on the tongue, using the SwissModel platform and validated its stereochemistry with Ramachandran plots, which placed 85.5 percent of residues in the most favored regions. Docking all 22 peptides against the receptor produced binding energies from minus 9.5 to minus 6.3 kilocalories per mole, with lower values indicating more stable complexes. Five peptides stood out with binding energies below minus 8 kilocalories per mole: HAKIDAAEEEKY, GDEESYTVFK, EQAEEERYFRA, DPDEEALRRSR and EDEADDWARR, dubbed HY-12, GK-10, EA-11, DR-11 and ER-10 respectively.

The five peptides were chemically synthesized at greater than 95 percent purity and put to the test with human panels and an electronic tongue. Taste dilution analysis revealed umami thresholds of 0.0625 milligrams per milliliter for ER-10, 0.125 for HY-12 and DR-11, and 0.25 for GK-10 and EA-11, all below the 0.3 threshold of MSG itself. ER-10, a ten-residue peptide rich in acidic residues and featuring a tryptophan, delivered the strongest umami, described as intense savory with slight salty and sweet notes. When added to a 0.35 percent MSG solution, the peptides boosted umami scores by 8.82 to 36 percent, with ER-10 and DR-11 the most powerful enhancers. Intriguingly, the electronic tongue showed HY-12 producing the lowest initial umami but the strongest umami aftertaste, suggesting sustained receptor binding that human tasters, whose umami perception suppresses competing flavors, experienced differently. Fourier transform infrared spectroscopy showed all five peptides are dominated by beta-turn and random coil structures, flexible conformations that appear central to their taste activity.

The molecular simulations explained why. Docking revealed the peptides form 10 to 28 hydrogen bonds with T1R1/T1R3, making hydrogen bonding the dominant stabilizing force, supplemented by electrostatic and hydrophobic interactions. Eleven receptor residues showed high interaction frequency, including Asp108, Asn150, Gln221, Gln222 and Ser217, while energy decomposition identified Lys155, Gln52, Ser109, Ser216, Arg255, Ser217 and Met151 as the key anchoring sites. Within the peptides themselves, arginine, phenylalanine, tryptophan and lysine contributed most to binding, and ER-10’s tryptophan alone formed seven hydrophobic contacts, likely explaining its supremacy. Compared with typical short umami peptides, these long chains showed a broader, multi-site binding footprint touching both core and peripheral receptor residues. Docking against the alternative umami receptor mGluR4 yielded weaker binding energies, indicating T1R1/T1R3 is the primary driver of the peptides’ taste while mGluR4 plays a supporting role.

Hundred-nanosecond molecular dynamics simulations in AMBER confirmed the docking picture. Four of the five peptide-receptor complexes held steady with root mean square deviations between 4 and 6 angstroms, while EA-11 proved structurally unstable, matching its weakest sensory performance. MM/GBSA binding free energy calculations ranked ER-10 most tightly bound at minus 77.23 kilocalories per mole, with electrostatic energy as the chief favorable contributor. The authors argue their integrated strategy sidesteps the slow, costly traditional chromatography workflow and, crucially, extends computational umami prediction into the long-chain territory where existing models have faltered. Beyond illuminating why Dengchuan beef tastes so good, the findings point toward rationally designed, beef-derived umami peptides as natural, nutritious flavor enhancers that could reduce sodium reliance in processed foods, though the team notes that interactions with nucleotide enhancers like inosinate and guanylate still need to be untangled.

Subject of Research: Identification of novel long-chain umami peptides in Dengchuan beef and their molecular taste mechanism via the T1R1/T1R3 receptor.

Article Title: Analysis of novel umami peptides in Dengchuan beef and their taste mechanism: Integrated peptidomics, machine learning and molecular simulation studies

Article References: Zheng, W., Chai, Y., Wang, Y., Yang, X., He, J., Li, Q., Wei, G., Huang, A., & Li, Y. (2026). Analysis of novel umami peptides in Dengchuan beef and their taste mechanism: Integrated peptidomics, machine learning and molecular simulation studies. Current Research in Food Science, 13, Article 101570. https://doi.org/10.1016/j.crfs.2026.101570

Image Credits: AI Generated

DOI: 10.1016/j.crfs.2026.101570

Keywords: umami peptides, Dengchuan beef, peptidomics, machine learning, molecular docking, T1R1/T1R3 receptor, molecular dynamics simulation, taste threshold, flavor chemistry, food science, Analysis, novel