Combining correlation analysis and machine learning to identify indicators strongly linked to quality variations in low-salt fermented shrimp sauce during storage

Abstract

Developing reduced-salt shrimp sauce has become an inevitable trend, while increases the risk of spoilage during its storage. The current quality control criteria for shrimp sauce fails to provide accurate and holistic quality assessment. The establishment of new quality control standards is need and necessitates clarifying the quality evolution patterns of low-salt shrimp sauce. During storage, pH, total volatile basic nitrogen (TVB-N), histamine, tartaric acid, succinic acid, hypoxanthine and aromatic compounds increased significantly, while total acidity, inosine monophosphate, alcohols, and aldehydes decreased. Correlation analysis and machine learning found that the level of histamine, total acidity and TVB-N could more accurately reflect shrimp sauce spoilage. Moreover, acetic acid and isoleucine were closely associated with flavor deterioration, while nonanal was strongly linked to odor decline. These findings clarified the relationship between physicochemical changes and flavor evolution, providing a foundation for establishing a robust and enhanced quality evaluation system for low-salt shrimp sauce.

Highlights

• •pH, TVB-N, histamine and tartaric acid were positively related to shrimp sauce deterioration.
• •Total acidity, IMP, alcohols and aldehydes negatively linked to shrimp sauce deterioration.
• •Histamine, total acidity and TVB-N were reliable markers of shrimp sauce spoilage.
• •Acetic acid, isoleucine and nonanal were key indicators reflecting quality deterioration.

1. Introduction

Shrimp sauce is a traditional aquatic condiment produced through the natural fermentation of fresh shrimp under high-salt condition, which owns distinctive flavor owing to a series of complex biochemical processes during fermentation, including protein hydrolysis, amino acid metabolism, lipid degradation, and microbial activity (El Sheikha & Hu, 2020). Traditionally, employing high salt content in shrimp sauce is a primary method for its preservation, effectively inhibiting microbial spoilage and extending shelf life. In recent years, the development of low-salt or reduced-salt formulations has become an inevitable trend in food market, driven by the increasing consumer awareness of health and nutrition (El Sheikha & Hu, 2020). However, the reduction of salt content inevitably accelerates microbial activity and biochemical reactions, raising significant challenges for ensuring the stability and safety of reduced-salt fermented seafood condiments during storage (El Sheikha & Montet, 2014).

Regulatory standards for aquatic products and processed aquatic products have been well established worldwide (Hazards et al., 2024). In contrast, quality control criteria for fermented seafood condiments such as shrimp sauce and fish sauce are relatively simple. In most cases, only compliance with sensory quality, basic microbiological safety requirements and broad limits for biogenic amines are mandated (Codex Alimentarius Commission, 2011; National Health and Family Planning Commission of the People's Republic of China, 2014). In reality, these condiments continue to undergo slow microbial activity, lipid oxidation reaction as well as non-enzymatic reactions, leading to dynamic changes in product quality (Ray, El Sheikha & Sasi Kumar, 2014). Therefore, the prevailing quality metrics fail to provide accurate and holistic quality assessment of shrimp sauce. The establishment of new quality control standards is need and necessitates clarifying the quality evolution patterns of low-salt fermented shrimp sauce during storage. However, a significant research gap exists regarding quality changes in low-salt shrimp sauce throughout storage periods.

In recent years, machine learning has emerged as a powerful tool for analyzing complex food quality data. By constructing predictive models based on input variables, these methods can aid in optimizing food processing techniques. Beyond processing, machine learning has also been applied in agriculture, food packaging, and food safety (Pandey et al., 2023). Importantly, these approaches can identify key predictive variables from high-dimensional datasets, offering a systematic and objective means to uncover the factors driving product quality deterioration and flavor changes. Additionally, integrating machine learning with traditional correlation analysis is a powerful tool using in evaluating and predicting product quality (Khorramifar et al., 2023). Therefore, this study applies such methods to shrimp sauce storage for offering a promising strategy to identify direct physicochemical markers of deterioration and key flavor compounds driving off-flavor formation.

Previous studies have mainly focused on fermentation processes and quality changes during fermentation, while the storage stage after fermentation has received limited attention. In addition, most research has been conducted on high-salt fermented products, whereas low-salt fermented seasonings, which align with current health trends, face greater challenges in maintaining quality during storage. Therefore, identifying key indicators and flavor compounds associated with quality and flavor deterioration in low-salt shrimp sauce during storage is essential for ensuring product quality stability.

This study aimed to systematically investigate the quality deterioration of shrimp sauce during storage, then focuses on identifying physicochemical indicators that serve as direct markers of quality decline and key flavor compounds driving off-flavor formation, which can provide reference for developing quality control standards for low-salt shrimp sauce. Specifically, representative physicochemical indices and flavor compounds across different storage stages were monitored. Furthermore, correlation analysis combined with machine learning was employed to explore and screen multidimensional data, with the aims of identifying reliable quality markers for evaluating and predicting storage stability, and elucidating the flavor compounds most strongly associated with deterioration. The findings are expected to provide a scientific foundation for establishing quality control standards, supporting the development of reduced-salt shrimp sauce products, and promoting the standardization, modernization, and shelf-life management of fermented seafood condiments.

2. Materials and methods

2.1. Materials

All Antarctic krill used in this study were sourced from China National Fisheries (Beijing, China), harvested in 2024, graded M+, and stored at −20 °C.

2.2. Preparation, sterilization, and preservation of Antarctic krill shrimp sauce

The brewing method of Antarctic krill shrimp sauce was based on the procedure described by Qu, Wang, Liu, et al. (2025). Briefly, the material-to-liquid ratio was adjusted to 0.9, and the pH was set to 7.5. Enzymatic hydrolysis was conducted at 43 °C for 10 h. Fermentation was performed through stepwise salt addition and temperature increase: first, 3% salt was added, and fermentation was carried out at 15 °C for 3 days; then the salt concentration was increased to 10% and fermentation continued at 35 °C for 10 days, followed by an additional fermentation step at 42 °C for 2 days. The final fermented product was centrifuged to collect the supernatant, which constituted the Antarctic krill shrimp sauce.

After sterilization at 85 °C for 30 min, the shrimp sauce was aseptically filled into sterile PET bottles and stored at room temperature (25 °C, 45% relative humidity) for six months. Samples were collected monthly and labeled as M0, M1, M2, M3, M4, M5, and M6. All samples were stored at −20 °C prior to analysis.

2.3. Determination of basic physical and chemical indicators

2.3.1. Total viable count (TVC) determination

The plate count agar (PCA) pour plate method was used to determine the total viable count of shrimp sauce during storage according to international standards (ISO 4833-1, 2013). Under sterile conditions, 1 mL of the homogenized sample was transferred to a sterile Petri dish. Then, 15–20 mL of PCA, cooled to approximately 46 °C, was poured into the dish and gently swirled to mix. A blank control was prepared by adding 1 mL of sterile saline to a separate plate. All plates were incubated at 37 °C for 48 h, after which colonies were counted. Results were expressed as total colony counts (CFU/mL).

2.3.2. Color, pH and acidity

The pH was measured using an acidometer (FiveEasy Plus, Mettler Toledo, Switzerland) according to the method of McCrudden (1922). A 10 mL sample was mixed thoroughly with 90 mL of distilled water. The calibrated pH probe was then immersed below the surface of the mixture, and the pH value was recorded once the reading stabilized.

Acidity was determined by titration according to the method of Qu, Wang, Kong, et al. (2025). Briefly, 5 mL of sample was mixed with 45 mL of distilled water and titrated to pH 8.2 with 0.1 M NaOH. The volume of NaOH consumed was recorded, and the acidity was calculated.

The color of the sample was measured using a CR-200 spectrophotometer (Minolta Camera Co., Osaka, Japan) according to the national standard method (AQSIQ, 2002) with minor modifications. A 2 mL aliquot of the well-mixed sample was transferred to a sample cell, and the calibrated colorimeter probe was placed against the cell under complete light-shielding conditions. The color parameters were measured and recorded accordingly.

2.3.3. Total protein (TP), total volatile basic nitrogen (TVB-N) and amino acid nitrogen (AAN)

The contents of TP and TVB-N were determined using the Kjeldahl method, following the procedure described by Qu, Wang, Liu, et al. (2025). Briefly, 1 mL of the mixed sample was placed in a digestion tube, followed by the addition of a catalyst (copper sulfate and potassium sulfate) and 10 mL of sulfuric acid. The mixture was digested at 420 °C. After digestion, the distillate was absorbed with boric acid and titrated with a standard hydrochloric acid solution. Protein content was calculated based on the volume of acid consumed. For TVB-N determination, 5 mL of the sample was mixed with 75 mL of distilled water and allowed to stand at room temperature for 30 min. After adding 1 g of magnesium oxide, the mixture was immediately subjected to distillation and titration.

AAN was determined by colorimetric methods based on AOAC Official Method 920.87 (AOAC International, 2016) with minor modifications. The colorimetric reagent was prepared by mixing 7.8 mL acetylacetone and 15 mL 37% formaldehyde and diluting to 100 mL with deionized water. The sample was mixed with a sodium acetate–acetic acid buffer (pH 4.8) and the color reagent, then heated in a water bath at 100 °C to induce a color reaction that produced a yellow derivative. The absorbance was measured at 400 nm to calculate the AAN content.

2.3.4. Biogenic amines (BA)

The content of BA was determined by high-performance liquid chromatography (HPLC), following the method described by Abre et al. (2023). Briefly, the sample was mixed with 100 mg/L 1,7-diaminoheptane (internal standard), saturated sodium bicarbonate solution, 1 M NaOH, and 100 mg/mL dansyl chloride (derivatization reagent), and derivatized for 15 min in a water bath at 60 °C. The derivatized sample was separated on a C18 column (ZORBAX Eclipse XDB-C18, 4.6 mm × 150 mm, 3 μm) using an Agilent 1260 HPLC system (Agilent Technologies Inc., CA, USA) equipped with a UV detector. The detection wavelength was set at 254 nm, and the mobile phase consisted of acetonitrile and 0.01 M ammonium acetate solution containing 0.1% acetic acid.

2.4. Non-volatile metabolites

2.4.1. Free amino acids (FAA)

The FAA content was determined using an automatic amino acid analyzer (Hitachi L-8900, Japan) equipped with a post-column ninhydrin derivatization system (PF system), following the method of Qu et al. (2025). Briefly, samples were extracted with 0.02 M HCl and centrifuged to remove impurities. Then, 5% sulfosalicylic acid was added, followed by refrigerated centrifugation. The resulting supernatant was filtered through a membrane before analysis.

2.4.2. Organic acids

The Organic acids content was determined following the method of Qu et al. (2025). Organic acids in the samples were extracted with 0.02 M disodium hydrogen phosphate, separated on a C18 column (Waters Atlantis C18, 4.6 mm × 150 mm, 3 μm), and detected using a high-performance liquid chromatograph (Agilent 1260, Agilent Technologies Inc., CA, USA) equipped with a UV detector. The mobile phase consisted of 0.02 M disodium hydrogen phosphate buffer (pH 3.0), and detection was performed at 210 nm.

2.4.3. Flavor nucleotides

The nucleotide content was determined following the method of Qu et al. (2025). Briefly, nucleotides were extracted from the sample using perchloric acid, and the pH was adjusted to 6.0 with NaOH. After precipitation and removal of impurities, the sample was analyzed by HPLC using an Agilent 1260 system (Agilent Technologies Inc., CA, USA) equipped with a UV detector and a CAPCELL PAK C18 column (SG120, 4.6 mm × 150 mm, 5 μm). The mobile phase consisted of a 1:1 (v/v) mixture of 0.02 M KH₂PO₄ and 0.02 M K₂HPO₄ buffer at pH 6.0, and detection was performed at 254 nm.

2.5. Volatile flavor substances

Volatile flavor compounds were determined using solid-phase microextraction (SPME) coupled with triple quadrupole gas chromatography–mass spectrometry (GC–MS/MS), following the method of Qu et al. (2025). Briefly, 2 g of sample was placed in a 20 mL headspace vial, and 5 μL of 2,4,6-trimethylpyridine (20 μg/mL) was added as an internal standard. The vial was preheated at 55 °C for 5 min, followed by extraction using a 50/30 μm DVB/CAR/PDMS fiber (Supelco, Bellefonte, PA, USA) at 55 °C and 550 rpm for 30 min. Analysis was conducted on a GC–MS/MS system (Agilent GC8890–MS/MS 7000D, Agilent Technologies Inc., CA, USA) equipped with a DB-5MS capillary column (30 m × 0.25 mm × 0.25 μm, Agilent J&W, CA, USA). The oven temperature program was as follows: initial temperature at 40 °C, ramped to 90 °C at 4 °C/min and held for 2 min, then increased to 200 °C at 5 °C/min and held for 5 min, and finally raised to 240 °C at 10 °C/min. Volatile compounds were identified by comparing mass spectra with entries in the WILEY 32 and NIST 17.L libraries.

2.6. Calculation of taste activity value (TAV) and odor activity value (OAV) values

TAV and OAV were calculated as the ratio of the compound concentration to its respective taste or odor threshold, as shown in Eqs. (1), (2).

$$\mathrm{TAV}=\frac{{\mathrm{C}}_{\mathrm{nv}}}{{\mathrm{T}}_{\mathrm{nv}}}$$ (1)

$$\mathrm{OAV}=\frac{{\mathrm{C}}_{\mathrm{V}}}{{\mathrm{T}}_{\mathrm{V}}}$$ (2)

C_nv represented the concentration of taste substances (mg/100 g), and T_nv was the taste threshold of corresponding taste substances (mg/100 g); C_V represented the concentration of volatiles (μg/kg); T_V was the odor threshold of corresponding volatiles (μg/kg).

2.7. Electronic tongue (e-tongue)

The macrotaste profile of the samples was analyzed using an e-tongue system (Insent SA402B Plus-EX, Japan) equipped with six taste sensors (AAE: umami, CT0: salty, CAO: sour, C00: bitter, AE1: astringent, GL1: sweet) and three sensing electrodes following the method described by Toko (2000). The positive electrode cleaning solution consisted of 10 mM KOH, 100 mM KCl, and 30% ethanol, while the negative electrode cleaning solution contained 100 mM HCl and 30% ethanol. The reference solution was prepared with 30 mM KCl and 0.3 mM tartaric acid. Samples were diluted tenfold, dispensed into sample cups, and analyzed in quadruplicate. The last three measurements with stable readings were used for data analysis.

2.8. Sensory evaluation

Sensory evaluation was carried out according to the method described by Qu et al. (2025) using a panel of 20 professionally trained panelists (10 males and 10 females, aged 22–32 years) in a well-lit, ventilated, and odor-free environment at 25 °C and 65% relative humidity. For each evaluation, 20 mL of the mixed sample was placed in a transparent plastic cup. Panelists observed the color for 30 s and recorded the color score. Then, using a straw, two drops of the sample were placed on the tongue, and the taste was evaluated for 30 s. For aroma assessment, the sample was preheated to 55 °C, sniffed for 30 s, and the aroma score was recorded. Samples were presented in random order. The scoring criteria and details of the shrimp sauce were shown in Table 1. After each evaluation, panelists rinsed their mouths with warm water and rested for 30 min.

Table 1.

Scoring standards for the sensory evaluation of shrimp sauce (National Health and Family Planning Commission of China, 2014).

Attribute	Score (points)
	10	5	1
Color and transparency	Dark brown/reddish brown/orange red, clear and transparent, without impurities	Light color but transparent, or slightly turbid	Turbidity, sedimentation or obvious suspended matter
Odor	Umami and fragrant, no fishy or peculiar smell	Light odor or slight odor	Rancid, pungent or obvious fishy smell
Taste	Harmonious taste, strong umami flavor, mellow and nonirritating	Too salty/light, lack of umami, slightly sour/bitter/unpleasant taste	Sour/bitter/unpleasant taste
Overall satisfaction	Well-balanced flavor, moderate saltiness and umami, rich and layered taste, highly consistent with expectations or traditional shrimp sauce characteristics, stimulates appetite and demonstrates strong consumer appeal	Acceptable taste, though somewhat bland or unbalanced, minor off-flavors or imbalances may be present but do not significantly affect edibility	Seriously unbalanced flavor, noticeable off-odors or unpleasant sensory characteristics, unacceptable for consumption and fails to meet basic product standards

All participants were informed that they were completely voluntary and free to do this experiment, and informed consent was obtained from them in full accordance with the 1975 Declaration of Helsinki. Ethical approval for the involvement of human subjects in this study was granted by the Ocean University of China Research Ethics Committee, reference number OUC-HM-2025-043.

2.9. Correlation analysis and machine learning

To identify key quality parameters closely associated with spoilage indicators during the storage of Antarctic krill sauce, Pearson and Spearman correlation analysis was first conducted to examine the linear relationships between each individual variable and both storage time and spoilage indicators. In addition, machine learning regression models were employed for multivariate feature selection and importance evaluation, including LASSO regression and Random Forest Regressor. LASSO was used for sparse feature selection, while Random Forest was applied to capture non-linearities and interaction effects. To verify the significance of the variables, we evaluated them using correlation coefficients (r-values) and model-derived importance scores. First, variables with r ≥ 0.8 in Pearson and Spearman correlation analyses were selected. Next, Lasso and Random Forest models were applied, and variables exhibiting high importance values in both models were retained for further analysis.

2.10. Data analysis

To ensure data accuracy, all measurements were performed in triplicate. Statistical analyses were conducted using SPSS (SPSS Inc., Chicago, IL, USA), and results are expressed as mean ± standard deviation (SD). Differences between groups were analyzed by one-way analysis of variance (ANOVA). Bar charts, heatmaps, and other visualizations were generated using Origin 2021 (OriginLab, Northampton, MA, USA).

3. Results and discussion

3.1. Basic indicators

3.1.1. Microbial content, color, pH, and acidity

The determination of total colony count, color difference, pH, and acidity are of great significance for evaluating the quality, safety, and shelf life of shrimp sauce. Fig. 1a illustrates the changes in total colony count of shrimp sauce during storage. TVC is an important indicator of a product's hygienic quality and microbial stability, with excessive counts posing a risk of spoilage and food safety hazards. In this study, microorganisms were first detected in shrimp sauce after three months of storage and increased markedly thereafter, reaching 12 CFU/mL by six months of storage. Color difference, as a visual indicator of quality, is an important factor influencing the sensory perception and consumer acceptance of a product. Figs. 1b-d depict the changes in L*, a*, and b* values of shrimp sauce during storage. At the end of fermentation, the L*, a*, and b* values were 38.69, 0.20, and − 1.81, respectively. During storage, all three-color parameters exhibited a trend of initially decreasing, followed by an increase, and then a subsequent decrease. Notably, in the later stages of storage, the L* value (after M5) and the b* value (after M3) showed no significant changes, remaining approximately at 38.45 and − 2.14, respectively. These results suggested that the color of shrimp sauce darkened during the early storage period (shifting toward brown and bluish-green), then brightened, and ultimately darkened again over time. pH and total acidity are key physicochemical parameters that influence microbial growth, enzymatic activity, and flavor development. They also reflect the endogenous enzyme activity and microbial metabolism within the sample. As shown in Fig. 1e, during storage, the pH of shrimp sauce increased significantly, accompanied by a marked decrease in total acidity. Specifically, the pH rose from an initial value of 6.28 to 7.39, while the total acidity decreased from 1.53 to 0.49.

Fig. 1.

Changes in basic quality indicators of shrimp sauce during storage. (a) Total viable count; (b) L* value; (c) a* value; (d) b* value; (e) pH and acidity. Different letters indicate significant differences (p < 0.05).

Studying the changes in total bacterial count, color, pH, and acidity during the storage of shrimp sauce provided comprehensive insight into the dynamic variations of its sensory quality and physicochemical properties. Pasteurization effectively inhibited most microbial growth and enzymatic activities in the product. Therefore, the observed color changes during storage are likely attributed to non-enzymatic browning reactions (such as the Maillard reaction), oxidation processes, pigment degradation. In the study by Sunds et al. (2018), the L*, a*, and b* values of UHT-sterilized milk changed under different storage conditions. These changes were attributed to the Maillard reaction, which ultimately led to the formation of melanoidins. The pH and acid value of the samples changed during storage, indicating that pasteurization did not completely inactivate all microorganisms and enzymes. During storage, protein and amino acid degradation, microbial metabolism, and other biochemical reactions occurred, leading to the production of amines or the consumption of acidic substances. As a result, the pH gradually increased, which indirectly caused a reduction in the acid value of the samples. These findings are consistent with the study of pH changes in fish during storage reported by Joung and Min (2018).

3.1.2. Nitrogen-related indicators

The contents of TP and AAN are key indicators for evaluating the quality of shrimp sauce. They reflect the extent of protein degradation and are closely associated with the product's nutritional value and sensory quality. TVB-N is an important parameter for assessing the freshness and overall quality of shrimp sauce, as it relates to the metabolic activity of spoilage bacteria and endogenous enzymes (Gao et al., 2023). Figs. 2a-c illustrate the changes in TP, AAN, and TVB-N contents of shrimp sauce during storage. The TP content remained relatively stable throughout the storage period, ranging from 12.92% to 14.78%, with only a slight decrease observed after six months. Similarly, the AAN content remained relatively constant (0.94–1.09 g/100 mL), although a significant decline was noted at month 3 (M3), followed by a slight increase. In contrast, the TVB-N content showed a significant upward trend over time, increasing from an initial value of 92.85 mg N/100 mL to 199.42 mg N/100 mL by the end of the storage period. In summary, no extensive degradation or accumulation of proteins and amino acids occurred during the storage of shrimp sauce. However, small-molecule alkaline nitrogenous compounds continued to accumulate. This phenomenon may be attributed to the activity of salt-tolerant and heat-resistant microorganisms that survived pasteurization. These microorganisms may have slowly decomposed small-molecule nitrogen sources and produced volatile amines over the course of storage.

Fig. 2.

Changes in nitrogen-related quality indicators of shrimp sauce during storage. (a) Total protein content; (b) Amino acid nitrogen content; (c) Total volatile basic nitrogen content; (d) Histamine content; (e) Cadaverine content; (f) Spermidine content. Different letters indicate significant differences (p < 0.05).

3.1.3. Changes in BA content during storage

BAs are metabolic products of spoilage and fermentation processes, primarily formed through the decarboxylation of amino acids by microbial decarboxylases. Their presence reflects microbial activity and is closely associated with product safety. The separation of the target biogenic amines and the internal standard was efficiently achieved, as shown in the representative HPLC chromatogram (Fig. S1). During the storage of shrimp sauce, three biogenic amines were detected: histamine, cadaverine, and spermidine (Figs. 2d-f). Histamine levels increased significantly over the course of storage, rising from 3.00 mg/L to 11.80 mg/L. Cadaverine levels remained relatively constant at approximately 0.75 mg/L during the early storage period (M0-M3), followed by a sharp decline to 0.38 mg/L at M5, and then a significant increase to 0.78 mg/L at M6. Spermidine was not detected at the beginning of storage but increased significantly over time, reaching 0.31 mg/L at M5 before declining to 0.24 mg/L at M6. Histamine, putrescine, and cadaverine are commonly used as quality indicators for fresh aquatic products, as their concentrations increase progressively with spoilage (Sivamaruthi, Kesika & Chaiyasut, 2021). In addition, previous studies have reported that biogenic amines such as histamine, tyramine, and phenylethylamine are often present at high levels in fermented aquatic products, reflecting intense microbial activity and amino acid decarboxylation during fermentation (Sivamaruthi, Kesika & Chaiyasut, 2021). Significantly elevated histamine levels indicate the presence of histidine decarboxylase-producing bacteria. These may include salt- and heat-tolerant strains, such as certain members of the Enterobacteriaceae family or Lactobacillus species (Wang, Ren, Wang, Bai & Li, 2015), which can survive pasteurization and remain metabolically active during storage. The accumulation of histamine is particularly concerning, as excessive intake is associated with potential food safety risks, including scombroid poisoning (Tao et al., 2025). Although clear international limits have been established for histamine, the levels detected in shrimp sauce after six months of storage remained well below these thresholds. For example, the U.S. Food and Drug Administration (FDA) has set 50 ppm of histamine as the guidance level for fish spoilage, and 500 ppm as the action level indicating a potential risk to human health (Saha Turna et al., 2024). The observed changes in cadaverine and spermidine levels may be attributed to the decomposition of amino acids or proteins, oxidative degradation of the amines, or microbial activity. Specifically, variations in the abundance or activity of microorganisms capable of lysine decarboxylation and arginine metabolism may have contributed to these fluctuations.

3.2. Sensory attributes

3.2.1. E-tongue analysis

As an objective, rapid, and highly reproducible tool for taste analysis, e-tongue technology minimizes human bias and subjective variability compared to traditional sensory evaluation. It generates stable and quantifiable data, providing a scientific basis for food flavor research. In this study, the e-tongue employed nine types of taste sensors. During the storage of shrimp sauce, only five taste attributes showed sensor response values greater than zero. These changes are summarized in Table 2. Across all groups, the response values for umami and saltiness were consistently high, indicating that these were the core taste attributes of shrimp sauce. Richness and bitterness also exhibited relatively high response values, suggesting notable contributions to the overall taste profile. In contrast, the response value for aftertaste-B remained below 1, indicating a minimal impact on flavor perception. During storage, the response values for umami and saltiness displayed similar wave-like trends, with peak values observed at M6 (14.56) and M3 (13.54), respectively. Richness initially decreased and then increased, reaching its maximum value of 5.06 at M6. Bitterness followed a pattern of increase–decrease–increase, peaking at 4.10 in M6. Although aftertaste-B started low, its response value increased significantly over time. Overall, the response values for umami, saltiness, richness, bitterness, and aftertaste-B exhibited an upward trend with prolonged storage, with most reaching their highest levels at M6. These results suggest that the accumulation and transformation of flavor substances during storage enhanced the intensity of multiple taste attributes.

Table 2.

The response values of shrimp sauce analyzed by electronic tongue.

Taste sensors	Response value
	M0	M1	M2	M3	M4	M5	M6
Umami	14.30 ± 0.01bc	14.30 ± 0.00bc	14.19 ± 0.10cd	14.43 ± 0.08ab	14.10 ± 0.09d	14.20 ± 0.08cd	14.56 ± 0.06a
Richness	4.58 ± 0.03b	4.59 ± 0.00b	4.31 ± 0.15c	4.68 ± 0.21b	4.13 ± 0.08c	4.62 ± 0.18b	5.06 ± 0.09a
Saltiness	12.71 ± 0.03de	13.03 ± 0.00c	12.64 ± 0.06ef	13.54 ± 0.02a	12.61 ± 0.04f	12.80 ± 0.07d	13.15 ± 0.10b
Bitterness	3.79 ± 0.03f	3.87 ± 0.00e	3.90 ± 0.02de	4.04 ± 0.01b	3.92 ± 0.02d	3.97 ± 0.02c	4.10 ± 0.01a
Aftertaste-B	0.45 ± 0.03e	0.50 ± 0.00de	0.54 ± 0.02cd	0.58 ± 0.01c	0.56 ± 0.03cd	0.66 ± 0.06b	0.77 ± 0.03a

The values are mean ± standard deviation (n = 3), different letters in the same row indicate significant differences (p < 0.05).

M0, M1, M2, M3, M4, M5 and M6 represent the storage time of 0, 1, 2, 3, 4, 5 and 6 months, respectively.

3.2.2. Sensory evaluation

As one of the most direct and comprehensive approaches for food quality control and flavor analysis, sensory evaluation plays a vital role in product development, process optimization, and the assessment of consumer acceptance. Table 3 presents the changes in sensory scores of shrimp sauce during storage. The results showed that the color and transparency scores of the shrimp sauce remained stable throughout the storage period, maintaining an average value of approximately 8.29. This suggests that the color and appearance had minimal impact on sensory perception during storage. In contrast, the odor score declined significantly over time, dropping from 8.48 to 3.62, indicating a shift in aroma from a fresh, pleasant, and mildly fishy scent to a more pronounced sour and strong fishy odor. The taste and overall satisfaction scores exhibited a trend of significant increase followed by a marked decline. Notably, the taste score peaked at 8.67 at M4, then decreased sharply to 7.60 at M5, and further to 6.14 at M6. This pattern suggested that umami and kokumi flavors intensified during the early storage stages but were later masked by undesirable sour notes after five months. The overall satisfaction score was highest at M2 (8.82), then declined significantly to 3.21 by M6. This indicated that during early storage, the flavor was more balanced and the taste richer, making the shrimp sauce more appealing to consumers. However, after two months, flavor deterioration became apparent. By M6, the taste had become unbalanced and the odor more pronounced, rendering the product less acceptable.

Table 3.

The score of shrimp sauce analyzed by sensory evaluation.

Attribute	Score (points)
	M0	M1	M2	M3	M4	M5	M6
Color and transparency	8.17 ± 0.09a	8.23 ± 0.38a	8.53 ± 0.18a	8.28 ± 0.06a	8.46 ± 0.12a	8.20 ± 0.18a	8.19 ± 0.25a
Odor	8.48 ± 0.10a	8.45 ± 0.10a	7.94 ± 0.35b	7.30 ± 0.26c	6.62 ± 0.23d	5.11 ± 0.19e	3.62 ± 0.07f
Taste	7.16 ± 0.15d	7.75 ± 0.08bc	8.08 ± 0.15b	8.06 ± 0.18b	8.67 ± 0.22a	7.60 ± 0.27c	6.14 ± 0.15e
Overall satisfaction	8.10 ± 0.09d	8.32 ± 0.11c	8.82 ± 0.09a	8.66 ± 0.04b	6.61 ± 0.08e	5.08 ± 0.04f	3.21 ± 0.03g

The values are mean ± standard deviation (n = 3), different letters in the same row indicate significant differences (p < 0.05).

M0, M1, M2, M3, M4, M5 and M6 represent the storage time of 0, 1, 2, 3, 4, 5 and 6 months, respectively.

Based on sensory evaluation criteria, a sample is considered to have deteriorated if any individual score for color and transparency, taste, odor, or overall satisfaction falls below 5, indicating significant flavor deviation and unsuitability for sale or consumption. After six months of storage, the shrimp sauce sample's odor and overall satisfaction scores declined to 3.62 and 3.21, respectively—both below the acceptable threshold. These results demonstrate that the sample had undergone substantial sensory quality deterioration at this stage, rendering its flavor profile unacceptable to consumers.

3.3. Identification and selection of important indicators associated with shrimp sauce deterioration

3.3.1. Correlation analysis

Pearson correlation evaluates the linear relationship between variables, helping to identify features that are linearly dependent on the target variable. Spearman rank correlation captures nonlinear but consistent trends by examining the monotonic relationship between ranked variables, reducing the impact of outliers. To identify features closely associated with the deterioration of shrimp sauce quality during storage, this study employed a combination of Pearson correlation analysis and Spearman rank correlation analysis. This approach enables comprehensive screening and validation of representative, informative features from multiple perspectives. The results of the Pearson correlation analysis are presented in a correlation heatmap (Fig. 3a), where asterisks (*) denote statistically significant correlations and the size of the boxes within the grid reflects the magnitude of the correlation (r) values. The results showed that the TVC, TVB-N, b*, pH, total acid, histamine as well as spermidine content were significantly correlated with the storage time. Among these, TVB-N, pH and histamine, exhibited extremely significant positive correlations with the storage time. In addition, TVC, TVB-N and histamine showed significant correlations with the overall satisfaction of shrimp sauce, with the odor sensory score showing a particularly strong positive correlation. Fig. 3b presents a heatmap of feature correlations based on Spearman correlation analysis. TVC, TVB-N, pH, total acid, histamine, and spermidine content all showed significant correlations with storage time, generally consistent with the results of the Pearson correlation analysis. Notably, only TVC exhibited a significant negative correlation with overall satisfaction of shrimp sauce.

Fig. 3.

Correlation analysis between the basic quality indicators of shrimp sauce, storage time, and overall consumer satisfaction. (a) Heat map of Pearson correlation coefficients; (b) Heat map of Spearman correlation coefficients. * indicating significant differences (p < 0.05).

In summary, the three analytical methods revealed linear relationships between variables and verified the consistency of quality trends at a systemic level, providing a robust statistical foundation for mechanistic analysis of shrimp sauce storage quality changes and the identification of key indicators. All analyses identified multiple quality parameters—such as TVC, pH, TVB-N, total acidity, histamine, and spermidine—that were significantly correlated with both storage time and overall satisfaction, suggesting their potential as reference indicators. However, these parameters exhibited a high degree of concordance in their variation patterns, resulting in strong collinearity. Therefore, determining the critical indicators of shrimp sauce quality deterioration during storage necessitates a comprehensive, multifactorial evaluation strategy.

3.3.2. Machine learning

In the initial stage, Pearson correlation analysis, Spearman rank correlation and the Mantel test were employed to explore the relationships between various quality parameters of shrimp sauce and storage time, as well as overall sensory satisfaction. Several parameters exhibited strong correlations with storage time and overall sensory satisfaction. However, these parameters were often intercorrelated, leading to potential multicollinearity issues. Therefore, to identify the most informative and non-redundant indicators, machine learning-based feature selection methods (e.g., random forest, LASSO) were subsequently applied. This combination of traditional statistical analysis and machine learning allowed both linear and nonlinear associations to be considered, resulting in a more robust selection of key variables.

A random forest prediction model was employed to identify key indicators associated with the deterioration of shrimp sauce quality. Feature importance was used to quantify the relative contribution of each variable to the model's predictive performance (Table 4). For storage time, histamine exhibited the highest contribution (17.81%), underscoring its role as the most relevant indicator of temporal changes during shrimp sauce storage. TVB-N also demonstrated a considerable contribution (17.04%). With respect to overall consumer satisfaction, total acidity emerged as the most influential factor (20.55%), followed by histamine (16.40%). To complement these findings, lasso regression was further applied to screen basic quality indicators. This approach identified additional indicators that can represent both storage time and overall satisfaction. In this context, importance reflects the strength of the linear relationship (measured by the absolute coefficient value), while the direction of the relationship is determined by the coefficient sign (Table 4). In contrast to random forest, lasso regression does not assign importance values to all predictors. Rather, it performs variable selection via sparsity, retaining only those predictors most strongly associated with the response and estimating their corresponding effects. When screening features related to shrimp sauce storage time, lasso regression identified only histamine and TVB-N. For overall shrimp sauce satisfaction, however, only TVB-N was selected. Taken together, these findings suggested that histamine served as the most direct indicator of storage time, whereas total acidity was most strongly associated with consumer satisfaction. In addition, TVB-N appeared to play an important role in shrimp sauce deterioration during storage.

Table 4.

Importance values of quality indicators related to shrimp sauce quality changes during storage.

Feature	For storage time		For overall satisfaction
	Lasso_Importance	RandomForest_Importance	Lasso_Importance	RandomForest_Importance
Histamine	0.51	0.18	–	0.16
TVB-N	0.00	0.17	0.06	0.16
TVC	–	0.17	–	0.16
Total acid	–	0.16	–	0.21
pH	–	0.13	–	0.12
Spermidine	–	0.13	–	0.12
b*	–	0.07	–	0.07

3.4. Changes in non-volatile flavor compounds during storage

3.4.1. FAAs

Protein hydrolysis is one of the key factors influencing the final flavor profile of fermented condiments (Wang, Xia, Gao, Xu & Jiang, 2017). The measurement of FAAs provides a useful indicator for assessing the extent of protein degradation during storage (Sun, Zhang et al., 2020). Fig. 4a presents the changes in FAA content in shrimp sauce during storage. Overall, the total FAA content remained relatively stable throughout the storage period. Among the individual amino acids, lysine (Lys), arginine (Arg), and leucine (Leu) consistently exhibited high concentrations, averaging approximately 1149.98 mg/100 mL, 1162.62 mg/100 mL, and 1220.61 mg/100 mL, respectively. Fig. 4b illustrates the changes in the content and TAVSs of key FAAs in shrimp sauce during storage. Overall, the variations in amino acid content were not statistically significant, as indicated by overlapping significance levels across time points. Taking glutamate (Glu) as an example, although there was no significant difference in its content, the overall trend of Glu content was observed: it initially decreased from 711.66 mg/100 mL at M1 to 693.11 mg/100 mL at M3, then increased to 744.40 mg/100 mL at M5, and finally declined slightly to 729.17 mg/100 mL at M6. Furthermore, the changes in Ala and Lys contents followed a pattern similar to that of Glu, with levels fluctuating from 858.38 mg/100 mL to 873.12 mg/100 mL for Ala, and from 1169.29 mg/100 mL to 1128.10 mg/100 mL for Lys. Val, on the other hand, exhibited a distinct fluctuation pattern: it first increased from 718.30 mg/100 mL at M0 to 748.33 mg/100 mL at M1, then decreased to 698.94 mg/100 mL at M3, increased again to 786.25 mg/100 mL at M5, and finally declined to 744.77 mg/100 mL at M6. In addition, there was no significant change in Arg during storage. Table 5 presents the TAVs of FAAs in shrimp sauce and their changes during storage. Glu, Lys, and Arg exhibited high TAVs, highlighting their key contributions to shrimp sauce flavor. Their TAVs followed a pattern of initially decreasing, then increasing, and finally decreasing again, reaching peak values at M5 (24.81), M4 (24.09), and M5 (23.76), respectively. Additionally, Val and Ala also showed relatively high TAVs, suggesting their important roles in the overall flavor profile.

Fig. 4.

Changes in free amino acid contents of shrimp sauces during storage. (a) Heat map illustrating changes in all free amino acids. The differently colored blocks within the circular area indicate distinct flavor characteristics. (b) Bar chart showing changes in key free amino acids in shrimp sauce over the storage period. Different letters indicate significant differences (p < 0.05).

Table 5.

The TAV of organic acids and flavor nucleotides of shrimp sauces.

Taste compounds	TAV
	M0	M1	M2	M3	M4	M5	M6
Asp	3.11 ± 0.07ab	3.07 ± 0.12ab	2.94 ± 0.11b	2.95 ± 0.18b	3.24 ± 0.22ab	3.27 ± 0.13a	3.34 ± 0.15a
Glu	23.71 ± 0.26bc	23.63 ± 0.16bc	23.61 ± 0.44bc	23.10 ± 0.37c	23.97 ± 0.39abc	24.81 ± 0.40a	24.31 ± 0.79ab
Thr	2.17 ± 0.02a	2.16 ± 0.09a	2.22 ± 0.05a	2.18 ± 0.13a	2.16 ± 0.10a	2.15 ± 0.05a	2.12 ± 0.11a
Ser	3.04 ± 0.01a	3.02 ± 0.17a	2.96 ± 0.12a	2.95 ± 0.15a	3.09 ± 0.06a	2.97 ± 0.14a	3.09 ± 0.06a
Gly	3.75 ± 0.04ab	3.72 ± 0.27ab	3.65 ± 0.04abc	3.34 ± 0.17c	3.88 ± 0.26ab	4.01 ± 0.16a	3.61 ± 0.21bc
Ala	14.31 ± 0.30ab	14.15 ± 0.39b	14.09 ± 1.51b	14.39 ± 0.87ab	15.63 ± 0.74ab	15.92 ± 0.48a	14.55 ± 0.64ab
Lys	23.39 ± 0.50ab	22.83 ± 0.15ab	22.32 ± 0.60b	22.37 ± 0.64b	24.09 ± 0.82a	23.43 ± 1.06ab	22.56 ± 0.88b
Pro	2.529 ± 0.07a	2.49 ± 0.19a	2.55 ± 0.15a	2.53 ± 0.04a	2.59 ± 0.03a	2.60 ± 0.03a	2.56 ± 0.07a
Val	17.96 ± 0.27ab	18.71 ± 0.36a	18.35 ± 0.21abc	17.47 ± 0.51cd	19.43 ± 0.33bcd	19.66 ± 1.18abcd	18.62 ± 0.79d
Met	3.21 ± 0.03c	3.20 ± 0.01abc	3.16 ± 0.02bc	3.10 ± 0.03c	3.13 ± 0.05ab	3.14 ± 0.03a	3.09 ± 0.03abc
Arg	23.09 ± 0.43ab	23.02 ± 0.30bc	22.76 ± 0.48bc	22.99 ± 0.96c	24.32 ± 0.57bc	23.76 ± 1.28a	22.83 ± 0.78bc
Ile	8.29 ± 0.09a	8.18 ± 0.15a	8.13 ± 0.13a	7.92 ± 0.21a	8.08 ± 0.25a	8.80 ± 0.29a	8.20 ± 0.31a
Leu	6.44 ± 0.19a	6.39 ± 0.28a	6.47 ± 0.17a	6.25 ± 0.15a	6.44 ± 0.31a	6.66 ± 0.20a	6.31 ± 0.12a
Tyr	1.70 ± 0.01b	1.87 ± 0.09a	1.82 ± 0.03a	1.85 ± 0.05a	1.88 ± 0.08a	1.86 ± 0.06a	1.87 ± 0.02a
Phe	6.62 ± 0.07c	6.88 ± 0.06bc	6.85 ± 0.05bc	6.71 ± 0.25c	7.41 ± 0.28a	7.07 ± 0.14ab	6.80 ± 0.20bc
His	10.85 ± 0.26a	9.09 ± 0.27b	8.36 ± 0.10c	8.03 ± 0.09d	8.14 ± 0.11cd	7.17 ± 0.08e	6.70 ± 0.03f
Tartaric acid	73.00 ± 1.86c	81.98 ± 2.65c	78.88 ± 6.62c	81.21 ± 2.59c	144.13 ± 5.46b	211.21 ± 9.92a	211.32 ± 17.45a
Malic acid	7.323 ± 0.16d	7.25 ± 0.20d	7.95 ± 0.08c	8.37 ± 0.14b	8.60 ± 0.12b	9.06 ± 0.14a	8.32 ± 0.15b
Lactic acid	4.89 ± 0.04b	4.82 ± 0.11b	5.92 ± 0.22a	5.12 ± 0.32b	5.70 ± 0.04a	4.83 ± 0.25b	4.35 ± 0.11c
Acetic acid	3.08 ± 0.15a	2.89 ± 0.10a	3.02 ± 0.14a	2.22 ± 0.11c	2.50 ± 0.09b	3.02 ± 0.05a	3.08 ± 0.02a
Citric acid	104.83 ± 1.02d	107.37 ± 1.92cd	108.58 ± 1.21c	110.38 ± 2.71c	114.84 ± 1.15b	116.55 ± 1.38ab	119.84 ± 1.46a
Succinic acid	182.20 ± 2.63f	202.75 ± 7.17e	222.02 ± 7.08d	241.02 ± 3.47c	249.02 ± 7.06bc	258.65 ± 0.98b	274.80 ± 2.12a
GMP	1.70 ± 0.02a	1.66 ± 0.02c	1.66 ± 0.05bc	1.56 ± 0.06d	1.64 ± 0.03c	1.69 ± 0.02ab	1.64 ± 0.01c
IMP	0.29 ± 0.01a	0.28 ± 0.01b	0.26 ± 0.01c	0.25 ± 0.01d	0.23 ± 0.01e	0.21 ± 0.01f	0.20 ± 0.00g
AMP	2.87 ± 0.05c	2.97 ± 0.07abc	2.89 ± 0.06c	2.95 ± 0.10bc	3.05 ± 0.06ab	3.05 ± 0.04ab	3.10 ± 0.07a

The values are mean ± standard deviation (n = 3), different letters in the same row indicate significant differences (p < 0.05).

M0, M1, M2, M3, M4, M5 and M6 represent the storage time of 0, 1, 2, 3, 4, 5 and 6 months, respectively.

3.4.2. Organic acids

The content of organic acids not only reflects the degree of fermentation and microbial activity in the product but also significantly influences its flavor and safety. Therefore, it serves as an important physicochemical indicator for evaluating product quality. Fig. 5 shows the changes in organic acid content in shrimp sauce during storage. The levels of tartaric acid, citric acid, and succinic acid increased significantly over time, with tartaric acid exhibiting the most pronounced rise. During the first three months, tartaric acid content remained stable at approximately 122.32 mg/100 mL. However, from the fourth to the sixth month, its concentration increased markedly, rising from 216.19 mg/100 mL to 316.98 mg/100 mL. In addition, from the first month of storage onward, the contents of citric acid and succinic acid in shrimp sauce showed a significant increasing trend, rising from 4831.79 mg/100 mL and 2149.20 mg/100 mL to 5392.68 mg/100 mL and 2912.84 mg/100 mL, respectively. During storage, malic acid and acetic acid in shrimp sauce exhibited different trends: malic acid first increased then decreased, while acetic acid initially decreased followed by an increase. After six months, malic acid content increased significantly from 363.46 mg/100 mL (M0) to 412.73 mg/100 mL (M6), whereas acetic acid remained stable at 32.68 mg/100 mL (M6). Additionally, lactic acid content showed a fluctuating pattern—initially rising, then falling, rising again, and finally decreasing. Compared to the initial level, lactic acid at M6 was significantly lower, decreasing from 616.71 mg/100 mL (M0) to 548.46 mg/100 mL (M6). Table 5 presents the TAVs of six organic acids during storage. Among them, succinic acid, citric acid, and tartaric acid exhibited very high TAVs, highlighting their important roles in shaping the flavor of shrimp sauce. Their TAVs also increased significantly throughout storage, rising from 182.20, 104.83, and 73.00 to 274.80, 119.84, and 211.32, respectively. Changes in organic acid content during shrimp sauce storage may result from the ongoing metabolic activity of certain microorganisms, particularly halophiles and lactic acid bacteria. Yu et al. (2022) reported that rapid fermentation of shrimp paste using halophilic starters, especially Staphylococcus nepalensis JS11, significantly increased organic acid levels. Similarly, Ozogul et al. (2021) confirmed that inoculation with lactic acid bacteria promotes organic acid accumulation. For instance, L. acidophilus produced significant amounts of acetic acid in octopus infusion broth, while L. mesenteroides subsp. cremoris and L. delbrueckii subsp. lactis generated substantial levels of succinic acid. Organic acids act as both substrates and products in microbial metabolism, leading to dynamic changes in their concentrations (Herlina & Setiarto, 2024). These changes reflect ongoing fermentation and influence the sourness, umami, and overall sensory characteristics of the product.

Fig. 5.

Changes in non-volatile flavor compounds in shrimp sauce during storage. (a)–(f) Changes in organic acid contents; (g)–(k) Changes in flavor nucleotides. Different letters indicate significant differences (p < 0.05).

3.4.3. Flavor nucleotides

Changes in nucleotide content reflect the extent of nucleotide degradation in a sample and are closely associated with both freshness and flavor quality. Nucleotides are key non-volatile flavor compounds in shrimp sauce, contributing significantly to its umami taste. Investigating the changes in nucleotide content during storage can help elucidate the accumulation and degradation patterns of umami-enhancing substances over time. As shown in the representative HPLC chromatogram (Fig. S2), the target nucleotides were well-separated, ensuring the reliability of the quantitative results. Figs. 5g-k illustrate the changes in nucleotide content in shrimp sauce during storage. The IMP content decreased significantly over time, from 7.26 mg/100 mL at the beginning to 4.92 mg/100 mL. In contrast, the HX content increased markedly, from 86.84 mg/100 mL to 120.68 mg/100 mL. Notably, GMP, AMP, and HXR exhibited overlapping trends during storage. Overall, AMP and HXR showed slight increases, while GMP exhibited a slight decline. Specifically, after six months of storage, GMP content decreased significantly from 21.22 mg/100 mL to 20.51 mg/100 mL, whereas AMP content increased significantly from 143.43 mg/100 mL to 155.20 mg/100 mL. Table 5 presents the changes in the TAVs of the three flavor nucleotides during shrimp sauce storage. Among them, AMP exhibited the highest TAV, indicating that it was the most important flavor-contributing nucleotide in shrimp sauce. Additionally, GMP also played a significant role in flavor development. The content of flavor nucleotides is closely associated with their degradation and the secondary metabolic activities of microorganisms (Sun, Gao, et al., 2020). There are two main metabolic pathways involved in nucleotide degradation: (1) ATP → ADP → AMP → IMP → HxR → Hx; (2) ATP → ADP → AMP → AdR → HxR → Hx. Zhu et al. (2021) found that the concentration of 5′-AMP was significantly higher than that of 5′-IMP during fish sauce fermentation, suggesting that pathway (2) is the primary route for ATP degradation, while pathway (1) plays a secondary role Furthermore, research by Gao et al. (2023) confirmed that AMP is the predominant 5′-nucleotide and the primary contributor to flavor among 5′-nucleotides in seafood (Zhang, Wang, Li & Liu, 2019). In crustaceans, in particular, the activity of 5′-AMP deaminase is low, resulting in the significant accumulation of 5′-AMP (Chen, Zeng, Rong & Lou, 2021; Chen & Zhang, 2007). At low concentrations, AMP contributes to sweetness and enhances umami and overall flavor complexity (Chen & Zhang, 2007). Yu et al. (2022) found that the 5′-AMP content in shrimp paste fermented with halophilic bacteria was significantly higher than that in traditionally fermented shrimp paste, suggesting that halophilic bacteria may inhibit the activity of AMP-degrading enzymes.

3.5. Changes in volatile flavor compounds during storage

Volatile flavor compounds are key contributors to the aroma and sensory quality of food. Studying their dynamic changes helps to elucidate the mechanisms of flavor formation and evolution. In addition, such studies support the evaluation of quality stability, guide process optimization and storage control, and ultimately enhance product quality and consumer acceptance. A representative total ion chromatogram (TIC) is shown in Fig. S3. The major volatile components were well-separated and identified using NIST 17.L and WILEY 32 libraries with high confidence. As an example, the mass spectrum and fragmentation pattern of decanal, a key contributor to the characteristic aroma of shrimp sauce, is provided in Fig. S3a. Fig. 6 illustrates the changes in the proportion of volatile flavor compounds in shrimp sauce during storage, as well as the changes in the OAVs of key flavor compounds (Qu, Wang, Liu, et al., 2025). As shown in Fig. 6a, aldehydes, ketones, aromatic compounds, and heterocyclic compounds were identified as the major volatile flavor constituents in shrimp sauce. The proportion of aldehydes decreased sharply over the storage period, dropping from 45.75% at the beginning to 9.87% at M6. In contrast, the proportions of aromatic and heterocyclic compounds increased significantly, rising from 24.5% and 8.14% to 59.31% and 15.44%, respectively, by M6. The proportion of ketones showed relatively minor fluctuations, increasing initially from 24.5% at M0 to a peak of 10.28% at M4. The proportions of alcohols, esters, and amines in shrimp sauce were relatively low. Notably, alcohols were no longer detected after M2 during storage. As shown in Fig. 6b, decanal and nonanal were the major contributors to the volatile flavor of shrimp sauce during the early stage of storage, with OAVs of 186.35 and 178.64 at M0, respectively. However, their OAVs declined significantly over time, decreasing to 66.73 and 30.86 by M6. Furthermore, the OAV of benzeneacetaldehyde exhibited a significant decreasing trend over the storage period, dropping from 106.92 at M0 to 17.35 at M6. In contrast, the OAVs of 5-methyl-3-heptanone and propanal showed a fluctuating, ripple-like pattern during storage—first decreasing, then increasing, and finally decreasing again. By M6, their OAVs had increased from 76.82 and 73.62 at M0 to 80.13 and 85.93, respectively. Notably, the OAV of propanal increased significantly during the later stages of storage. A comprehensive analysis of the OAV changes of these five key volatile compounds revealed a shift in dominant aroma contributors—from decanal and nonanal at the beginning (M0) to 5-methyl-3-heptanone and propanal by M6. During storage, the volatile flavor compounds in shrimp sauce may undergo oxidation, degradation, or formation reactions, resulting in either the enhancement or deterioration of flavor. These changes significantly influence its sensory characteristics, including umami, fishy, aromatic, and putrid odors. For instance, aldehydes such as nonanal and decanal, which initially contribute to fresh and pleasant aromas, may decrease over time due to oxidation, leading to a reduction in overall aroma quality. Xing et al. (2023) reported that aldehyde content in sausages tended to decrease during the fermentation process. This decrease was attributed to the oxidation of some aldehydes to acids, which subsequently reacted with alcohols to form esters. Meanwhile, the accumulation of compounds like trimethylamine or ammonia, produced by the degradation of nitrogenous substances, can introduce undesirable fishy or putrid odors. Additionally, the formation of heterocyclic compounds such as pyrazines or sulfur-containing compounds during Maillard or microbial reactions can enhance the umami and complexity of the sauce's flavor profile (Nie, Che, Wang, Huang & Qin, 2025).

Fig. 6.

Changes in volatile flavor compounds in shrimp sauce during storage. (a) Stacked bar chart showing the relative proportions of volatile flavor compounds; (b) Bar chart illustrating changes in key volatile flavor compounds during storage. Different letters indicate significant differences (p < 0.05).

3.6. Compounds associated with flavor deterioration in shrimp sauce

3.6.1. Screening of key deteriorating flavor substances

To identify the key flavor compounds associated with flavor deterioration of shrimp sauce during storage, a machine learning-based approach was employed. First, sensory scores for taste and odor were used as response variables, and Pearson and Spearman correlation analyses were conducted to screen both non-volatile and volatile flavor compounds with absolute correlation coefficients greater than 0.8. Subsequently, the selected compounds were further refined using Lasso regression and Random Forest analysis. The selected flavor compounds are summarized in Table 6. Among the non-volatile compounds, acetic acid, and isoleucine (Ile) were closely associated with flavor changes. Spearman correlation analysis indicated a strong relationship between acetic acid and flavor deterioration (r = −0.93), with importance values of 0.36 and 0.40 in Lasso regression and Random Forest models, respectively, suggesting that acetic acid was a key predictor of shrimp sauce flavor deterioration across both linear and nonlinear models. Additionally, Pearson correlation analysis showed that Ile was highly correlated with flavor changes (r = −0.84), with importance values of 0.50 and 0.33 in Lasso and Random Forest models, indicating a pronounced linear effect on flavor deterioration while also contributing to nonlinear predictive models. These findings demonstrated that both acetic acid and Ile were robust indicators of flavor deterioration during shrimp sauce storage. In the screening of volatile flavor compounds associated with odor changes during shrimp sauce storage, nonanal exhibited a strong correlation with flavor deterioration. Pearson and Spearman correlation analyses yielded coefficients of 0.90 and 1.00, respectively, indicating a highly linear and monotonic relationship. Moreover, nonanal showed importance values of 0.65 and 0.33 in Lasso regression and Random Forest models, respectively, suggesting that it was a significant predictor of odor changes in both linear and nonlinear models. These results highlighted nonanal as a key volatile flavor compound associated with odor deterioration during shrimp sauce storage.

Table 6.

The r and importance values of volatile compounds related to flavor deterioration during storage.

Feature		Pearson_r	Spearman_r	Lasso_Importance	RandomForest_Importance
Taste- nonvolatile compounds	Acetic acid	−0.67	−0.93	0.36	0.40
	Ile	−0.84	−0.54	0.50	0.33
Odor-volatile compounds	Nonanal	0.90	1.00	0.65	0.33

3.6.2. Possible metabolic mechanisms

The key flavor compounds responsible for the deterioration of shrimp sauce during storage primarily originate from specific metabolic pathways, including amino acid degradation, fatty acid oxidation, carbohydrate breakdown, and microbial fermentation. For example, Yuan et al. (2023) reported that acetic acid was produced via the acetyl-phosphate and pyruvate pathways of lactic acid bacteria. In addition, Ile can be generated through protein hydrolysis and amino acid metabolism. Branched-chain aldehydes and short-chain aldehydes primarily arose from branched-chain amino acids, whereas straight-chain aldehydes were mainly derived from fatty acids (Zeng, Xia, Jiang, Xu & Fan, 2017). Research by Van Ba et al. (2013) indicated that nonanal, an oxidation product of oleic acid, was associated with a green, greasy aroma. Additionally, Ding et al. (2020) reported a significant positive correlation between octanal levels in fish sauce and the oleic acid content of the raw fish. These findings suggested that during fermentation, oleic acid primarily underwent autooxidation to form 10-ROOH, which is subsequently converted into nonanal. These metabolic pathways contribute to the flavor profile of shrimp sauce, leading to characteristic changes such as increased sourness and the development of undesirable odors.

4. Conclusion

The quality variations of reduced-salt shrimp sauce during storage were systematically investigated, and the indicators and compounds strongly associating to quality decline were identified. During storage, microbial content, pH, TVB-N, histamine, and tartaric acid increased significantly, with TVB-N rising from 93.50 to 203.07 mg N/100 mL, histamine from 2.96 to 11.80 mg/L, and tartaric acid from 109.50 to 316.98 mg/100 mL, while acidity, IMP, alcohols, and aldehydes decreased (alcohols and aldehydes from 3.44% and 45.75% to 0% and 9.87%). Sensory evaluation showed improved harmony in early storage but the emergence of undesirable flavors later. After six months, shrimp sauce exhibited a pronounced off-odor, with total protein at 13.24%, AAN at 1.05 g/100 mL, TVB-N at 203.07 mg N/100 mL, pH 7.39, total acidity 7.39, and histamine 11.80 mg/L. Correlation analyses and machine learning identified histamine, acidity, and TVB-N as key spoilage markers, while acetic acid, Ile, and nonanal were strongly associated with flavor and odor deterioration. These findings clarified the relationship between physicochemical changes and flavor evolution, providing a foundation for establishing a robust and enhanced quality evaluation system for low-salt shrimp sauce. Recommendations for maintaining quality during storage include the implementation of a dynamic monitoring system focused on TVB-N, histamine and total acidity as early warning signals. Additionally, microbial activity during storage should be effectively controlled by optimizing pasteurization parameters and/or adopting appropriate cold-chain storage temperatures to suppress microbial metabolism. Furthermore, the biomarkers identified in this study offer a solid theoretical foundation and scientific support for the development of rapid detection technologies with improved sensitivity and accuracy, enabling efficient monitoring and evaluation of product quality in industrial practice.

CRediT authorship contribution statement

Wenhui Qu: Writing – original draft, Software, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Qing Kong: Resources, Project administration. Ruoshu Li: Methodology, Data curation. Weijia Liu: Investigation, Formal analysis. Yunqi Wen: Writing – review & editing, Supervision, Project administration, Methodology, Investigation, Funding acquisition, Conceptualization. Changhu Xue: Writing – review & editing, Supervision, Project administration, Funding acquisition, Conceptualization.

Ethics statement

The sensory evaluation was carried out in accordance with the China National Standards GB/T 29605–2013 (Sensory analysis - Guide for food sensory quality control), and the participants in the sensory study took part voluntarily. Samples were presented after the sterilization and were free from potential risks. The study contained no ethical issues and ethical approval was not required.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that influenced the work reported in this paper.

Appendix A. Supplementary data

Supplementary material

Data availability

Data will be made available on request.