The Development and Characterization of a Nervonic-Acid-Rich Structured Lipid

Abstract

Nervonic acid (NA), an ultra-long-chain monounsaturated fatty acid, has attracted widespread attention in recent years due to its notable neuroprotective and antioxidant effects. In this study, a structured lipid was developed by enzymatically interesterifying coconut oil, palm stearin, and NA. The effects of lipase type, reaction temperature, time, and enzyme dosage on NA incorporation were investigated. The highest NA content in the structured lipid was achieved under optimized conditions: a reaction temperature of 64.6 °C, a reaction time of 7.17 h, and an enzyme dosage of 8.46%. Subsequently, a machine learning model was constructed to predict the sliding melting point. The NA-rich structured lipid, designed and prepared for use as a plastic fat, comprised 59.34% unsaturated fatty acids (46.76% NA) and exhibited β’ crystal polymorphism. This combined experimental and computational approach is a reliable strategy for preparing functional structured lipids.

1. Introduction

With the evolving consumer demand for health-oriented and functional foods, the oil and fat industry is shifting its focus from traditional mass production to precision design. Structured lipids (SLs), engineered through enzymatic or genetic modifications of the triglyceride (TAG) backbone, represent an advanced class of lipids that enable the “packaging” of specific bioactive fatty acids [1,2]. This packaging enhances the stability, bioavailability, and processing compatibility of these fatty acids [3].

Nervonic acid (NA, C24:1, n-9), an ultra-long-chain monounsaturated fatty acid, has garnered significant attention due to its pivotal role in maintaining the health of brain white matter [4]. Notably, recent investigations have established that NA exhibits an exceptionally high Oxidative Stability Index (OSI), surpassing that of a wide range of other fatty acids. This characteristic holds significant practical implications for the development of functional foods, as it effectively addresses the pervasive challenges of lipid oxidation and rancidity that commonly hinder food product development. Specifically, a study by Haq et al. demonstrated that, under identical experimental conditions, the long-chain monounsaturated structure of NA (C24:1) confers a distinct oxidative stability advantage over both short-chain fatty acids and polyunsaturated fatty acids [5].

However, NA is present only in trace amounts in most natural oils (e.g., 6% in Acer truncatum). Although certain oilseeds such as Malania oleifera contain relatively high concentrations (55.7–67.0%), their large-scale industrial application remains hindered by multiple challenges [6]. Moreover, NA typically exists as a free fatty acid or in its ethyl ester form [7,8], forms that exhibit low absorption efficiency and may induce gastrointestinal discomfort [9]. Consequently, developing an efficient and precise method of “embedding” NA into the glycerol backbone, thereby creating structured lipids suitable for high-end applications, represents a critical research priority.

Microbial synthesis, particularly through metabolic engineering and CRISPR technologies represents a cutting-edge avenue for de novo production of high-value lipids, such as the long-chain polyunsaturated fatty acids DHA and ARA that are essential for infant nutrition [10,11]. Despite rapid advancements, the industrial deployment of this technology for complex structured lipids remains constrained by limitations regarding yield, cost and the intricacies of pathway engineering [12,13]. In this context, enzymatic modification, particularly transesterification, provides a well- established, highly selective, and environmentally friendly pathway for the precise delivery of NA [14,15]. This approach has been effectively demonstrated in prior research; for instance, enzymatic interesterification of oil blends was used to modify triacylglycerol profiles for margarine production [16], and solvent-free lipase-catalyzed acidolysis was used to precisely synthesize SLs with specific MLM structures [17].

In parallel, data-driven approaches, particularly machine learning (ML), serve as powerful strategies for accelerating the design of functional foods. Traditional means of developing plastic fats with specific melting profiles rely heavily on empirical, iterative methods. Conversely, ML models can uncover complex, non-linear relationships between composition and functional properties, enabling accurate prediction of thermal behavior [18,19]. Building on this capacity, we hypothesize that a fatty-acid-based supervised learning algorithm can effectively pre-screen novel NA-rich structured lipid formulations for suitable melting profiles, thereby streamlining the development process.

The aim of this study was to develop a functional, bioavailable NA-rich food ingredient. We prepared a high-NA-content margarine fat base via enzymatic transesterification of NA, coconut oil, and palm stearin, integrating an ML-assisted approach into this process. Response surface methodology was employed to optimize reaction conditions to maximize NA incorporation. A supervised learning algorithm was developed to predict sliding melting points and screen fractions suitable for margarine application. The resulting structured lipids were comprehensively characterized in terms of their fatty acid profiles, TAG composition, solid fat content (SFC), SMP, polymorphisms, and microstructures to assess their suitability for health-oriented margarine products.

2. Results and Discussion

2.1. Lipase Screening

The efficacy of three commercially immobilized lipases (Lipozyme TL IM, Novozyme 435, and Lipozyme RM IM) in catalyzing the incorporation of NA into a blend of CO and PS was evaluated. Reactions were conducted under the following preliminary conditions: a substrate weight ratio of CO:PS:NA = 1:1:1, a temperature of 60 °C, an enzyme loading of 10% (w/w, relative to total substrates), and a duration of 8 h in a solvent-free system. The incorporation of NA was quantified as its molar percentage in the final TAG products.

As illustrated in Figure 1a, the lipases exhibited significantly different catalytic efficiencies (p < 0.05). Lipozyme TL IM yielded the highest NA incorporation value, at 27.76%, followed by Novozyme 435 and Lipozyme RM IM. This performance hierarchy is likely due to the distinct regio-selectivity and acyl-transferase activity of Thermomyces lanuginosus lipase (Lipozyme TL IM), which appears to be particularly effective in facilitating the acidolysis involving the long-chain nervonic acid under the tested conditions. Based on this screening result, Lipozyme TL IM was selected for all subsequent experiments to optimize the reaction parameters for maximal NA incorporation.

2.2. Optimization of Reaction Conditions

2.2.1. Effect of Temperature

The impact of reaction temperature on NA incorporation is illustrated in Figure 1b. The NA content increased significantly as the temperature rose from 50 to 60 °C (p < 0.05), after which it plateaued. Further elevation to 70 °C did not yield a statistically significant enhancement (p > 0.05). This trend suggests that 60 °C approximates the optimal temperature for Lipozyme TL IM activity within this specific system, balancing substrate mobility and reaction kinetics while potential enzyme deactivation at higher temperatures. This observation aligns with previous reports on lipase-catalyzed esterification reactions [20]. Consequently, 60 °C was selected as the optimal temperature for subsequent experiments.

2.2.2. Effect of Time

Figure 1c depicts the time course of NA incorporation. The NA content increased progressively for up to 10 h, with the most pronounced increase occurring within the initial 6 h (p < 0.01). The incorporation rate decelerated significantly between 6 and 12 h, and a slight yet statistically significant decline was observed at 14 h. This decrease may be attributed to the onset of back-hydrolysis or acyl migration during prolonged incubation. Therefore, 6 h was identified as the optimal reaction time for maximizing yield while ensuring process efficiency.

2.2.3. Effect of Enzyme Load

The influence of Lipozyme TL IM dosage (6–14% w/w) on NA incorporation is depicted in Figure 1d. NA content increased markedly with enzyme load up to an increase of 8% (p < 0.001). However, further increments to 10% and 14% did not produce significant improvements (p > 0.05), indicating that substrate accessibility or reaction equilibrium became a limiting factor beyond this point. The highest incorporation level (28.14%) was achieved at an 8% enzyme load. This trend—wherein increasing enzyme concentration enhances incorporation up to a threshold, after which the benefit diminishes—aligns with established principles of enzymatic acidolysis and corroborates findings from similar studies employing various lipid substrates [21,22,23].

2.2.4. Response Surface Methodology

To systematically optimize NA incorporation, a response surface methodology (RSM) approach based on a Box–Behnken Design (BBD) was employed, with reaction temperature (X₁), time (X₂), and enzyme dosage (X₃) serving as independent variables and NA content (Y) serving as the response variable (Table S1, Supplementary Materials). A quadratic polynomial model (Equation (1)) effectively described the relationship between Y and the independent variables:

Y = 28.02 + 0.79X₁ + 0.94X₂ + 0.37X₃ − 0.19X₁X₂ − 0.04X₁X₃ − 0.57X₂X₃ − 0.77X₁² − 0.97X₂² − 0.77X₃²

(1)

ANOVA confirmed the model’s robustness (F = 58.09, p < 0.0001; Table S2). The model exhibited excellent predictive power, with a coefficient of determination (R²) of 0.9868 and an adjusted R² of 0.9698, indicating that over 98% of the response variability was captured. Additionally, the lack-of-fit test indicated non-significance (p > 0.05), further validating the model’s adequacy for prediction [24].

The model predicted a maximum NA content of 28.35% under the optimal conditions of 64.6 °C, 7.17 h, and an 8.46% (w/w) enzyme loading. Validation experiments conducted under these conditions yielded an average NA content of 28.24%, closely matching the prediction (relative error < 0.4%). This strong agreement confirms the RSM model is highly accurate and reliable. Compared with protocols requiring prolonged reaction times or excessive catalyst loads, the optimized conditions provide an efficient and scalable approach, highlighting the potential for industrial application. This study not only delineates the optimal technical parameters but also offers a validated predictive framework for tailoring structured lipid compositions via enzymatic interesterification.

Beyond the plant-derived purified NA enzymatic synthesis demonstrated here, alternative strategies include microbial synthesis and direct ester exchange using NA-rich oilseeds. For instance, metabolically engineered Yarrowia lipolytica has been employed for efficient NA production, with Wang et al. reporting a fermentation titer of 57.48 g/L, and Zhao et al. further enhancing NA yields by supplementing with auxiliary carbon sources such as rapeseed oil, showcasing the potential of microbial pathways [12,25]. While direct ester exchange simplifies the process and improves resource efficiency, complex trace components in crude oil may affect enzyme specificity and product uniformity. The experimental benchmarks and baseline data established in this study provide valuable references for future research on these more complex heterogeneous systems.

2.3. Development and Validation of a Machine Learning Model for SMP Prediction

2.3.1. Screening of Prediction Models

ML has emerged as a powerful tool for predicting the physical properties of food components, including the SMP. Building on previous successes in regard to predicting the SMP for ionic liquids using algorithms such as CatBoost (R² = 0.75) and deep learning models (R² = 0.90) [26], we developed and evaluated four ML models—GBDT, RF, KNN, and ANN—alongside a conventional MLR model. These models were trained exclusively on the fatty acid composition data to predict the SMP of nervonic-acid-structured lipids (NASLs).

A comparison of the performances of all six models is illustrated in Figure 2a–e, which plots predicted versus actual SMP values. Quantitatively, the models exhibited R² values ranging from 0.82 to 0.97 and root mean square error (RMSE) values from 0.86 °C to 2.14 °C. The ANN model consistently outperformed the others, achieving the highest R² (0.97) and the lowest RMSE (0.86 °C) on the test dataset (Figure 2e). This superior performance underscores the ANN’s ability to capture the complex, non-linear relationships between the multivariate fatty acid profile and the macroscopic melting point, a task for which the linear MLR model (R² = 0.82) was less adequate.

To validate the ANN model’s reliability and generalization, we analyzed learning curves and cross-validation metrics (Figure 3). Learning curves (Figure 3a) demonstrated that increasing training samples from 10 to 70 elevated both training and validation R² from 0.94 to 0.98, with minimal gap (<0.02), indicating robust generalization. Overfitting analysis (Figure 3b) further confirmed that the training–validation gap stabilized near 0.02 when samples ≥ 50, mitigating overfitting risks. 5-fold cross-validation (Figure 3c,d) revealed exceptional stability: mean R² = 0.9634, with most folds near 1. Deviation analysis (Figure 3e) identified only Fold 2 as an outlier (likely due to data partitioning anomalies), while others aligned closely, underscoring model robustness. These results collectively confirm the ANN’s high performance and generalizability, with minor fluctuations attributable to random data partitioning.

2.3.2. External Validation of Model Generalizability

An external validation set was employed to rigorously assess the practical applicability and robustness of the prediction models. This set comprised NASLs synthesized from three distinct substrate blends utilized in model training: PKO:PS:NA, CO:PO:NA, and CO:LO:NA. A comparison between the model-predicted and experimentally measured SMP values is presented in Figure 4.

The ANN model demonstrated exceptional generalizability, with an average prediction bias of less than 1% across all three novel lipid systems, closely aligning with the experimental measurements (e.g., a predicted value of 33.1 °C vs. an actual value of 33.3 °C for the CO:PO: NA blend). While the GBDT and KNN models also performed stably, maintaining prediction biases within 5%, the RF and MLR models exhibited significant volatility, with biases ranging from 0.6% to 18.7% and 1% to 34%, respectively. This successful external validation confirms that the ANN model is not only accurate but also reliably transferable to NASLs derived from oils with substantially different fatty acid compositions. This finding aligns with and extends previous work involving the use of ANNs in lipid engineering for process optimization [27,28], providing a powerful in silico tool for the rapid design of structured lipids with tailored melting properties.

2.4. Physicochemical Characterization of NASLs

Guided by the ANN predictions, six NASL variants with SMPs below body temperature (a key target for spreadable fats) and varying NA contents were selected for comprehensive characterization, linking their composition to functional properties.

2.4.1. Fatty Acid and Triacylglycerol Composition

The fatty acid profiles of the base oils and the synthesized NASLs are detailed in

Table 1. Fatty acid composition of neuronic-acid-structured lipids.

Fatty Acid	Coconut Oil	Palm Stearin	Nervonic Acid Structured Lipid Prepared from CO:PS:NA (%)
			1:1:3 (60%, NA)	2:2:5 (56%, NA)	4::4:9 (53%, NA)	5:5:8 (44%, NA)	2:2:3 (43%, NA)	4:4:5 (38%, NA)
C6:0	0.33	ND	0.10	0.12	0.14	0.15	0.15	0.16
C8:0	6.51	ND	1.40	1.57	1.78	1.94	2.05	2.25
C10:0	6.02	ND	1.37	1.49	1.60	1.81	1.88	1.97
C12:0	50.22	0.5	14.27	14.81	15.98	17.11	17.02	18.00
C14:0	18.67	1.07	4.16	4.53	4.75	5.42	5.62	5.91
C16:0	8.73	63.55	17.31	18.69	19.50	22.06	23.19	23.51
C18:0	3.36	4.86	2.02	2.20	2.23	2.61	2.57	2.76
C18:1	5.27	24.64	10.11	10.42	10.32	11.57	11.10	12.25
C18:2	0.82	4.91	2.09	2.11	2.13	2.40	2.23	2.49
C18:3	ND	0.07	ND	ND	0.02	0.02	ND	0.02
C20:0	0.06	0.31	0.15	0.16	0.15	0.17	0.16	0.16
C20:1	ND	0.09	0.03	0.02	ND	0.02	0.02	0.02
C22:1	ND	ND	0.24	0.35	0.21	0.18	0.27	0.25
C24:1	ND	ND	46.76	43.52	41.20	34.56	33.74	30.26
ΣMCFA	63.08	1.57	17.14	17.99	19.50	21.01	21.10	22.38
ΣSFA	93.90	70.29	40.66	43.57	46.13	51.27	52.64	54.72
ΣUSFA	6.10	29.71	59.34	56.43	53.87	48.73	47.36	45.28
Yield/NA	ND	ND	0.78	0.77	0.76	0.71	0.71	0.67

Note: “ND” stands for “Not Detected”; MCFA: Medium chain fatty acid; SFA: Saturated fatty acid; USFA: Unsaturated fatty acid; Yield = NA molar ratio in TAG/NASL molar ratio added; NASL: Nervonic acid structured lipid. All data are reported as mean values of triplicate analyses.

. Enzymatic transesterification was successfully used to engineer the lipid matrix. The NA content in NASL increased significantly, reaching up to 46.76% (1:1:3). Compared to the base oil CO and PS, this figure represents an enrichment in NA of about 10 times. The content of key palmitic acid (C16:0) decreased from 63.55% in palm stearin oil to 17.31% (1:1:3), indicating a significant desaturation trend, which may have potential significance in regard to reducing the risk of metabolic diseases [29]. The total USFA content increased to 59.34% (1:1:3), mainly due to the enrichment of NA (C24:1), which may endow the lipid with better physiological activity. Moreover, as the proportion of NA added to the raw material increased, the NA content in the final product increased correspondingly.

Restructuring at the molecular level was confirmed via TAG analysis (

Table 2. Possible nervonic-acid-containing triglyceride species in nervonic-acid-structured lipids prepared from CO:PS:NA.

NA-TAG a	Coconut Oil (CO, %)	Palm Stearin (PS, %)	Nervonic Acid Structured Lipid Prepared from CO:PS:NA (%)
			1:1:3 (60%, NA)	2:2:5 (56%, NA)	4:4:9 (53%, NA)	5:5:8 (44%, NA)	2:2:3 (43%, NA)	4:4:5 (38%, NA)
NANANA	ND	ND	1.81	1.05	1.44	0.85	0.80	0.56
LNANA	ND	ND	10.15	6.23	7.50	5.45	5.35	4.20
MNANA	ND	ND	10.65	8.23	7.30	4.97	5.20	3.92
MMNA	ND	ND	11.93	12.79	12.36	11.75	11.90	11.76
MLNA	ND	ND	27.87	26.97	26.47	22.26	22.77	20.02
LLNA	ND	ND	17.25	16.22	15.93	13.94	14.93	12.11
ΣNA-TAG	ND	ND	79.66	71.49	71.00	59.22	60.95	52.57

Note: ^a Abbreviations: NA-TAG, nervonic-acid-containing triglyceride species; NA, nervonic acid (C24:1, n-9); fatty acid codes: L, long-chain fatty acid (M, myristic acid, C14:0; P, palmitic acid, C16:0; S, stearic acid, C18:0; O, oleic acid, C18:1; L, linoleic acid, C18:2; and Ln, linolenic acid, C18:3); M, medium-chain fatty acid (Cy, caprylic acid, C8:0; Ca, capric acid, C10:0; and La, lauric acid, C12:0). Triacylglycerol (TAG) species are denoted without regard to fatty acid positional isomerism. For example, MMNA represents a monosubstituted TAG containing two medium-chain fatty acids (M) and one nervonic acid; LNANA represents a disubstituted TAG containing two nervonic acids and one long-chain fatty acid; and NANANA represents a trisubstituted TAG containing three nervonic acids. All data are reported as mean values of triplicate analyses.

). In the representative 1:1:3 NASL, NA-containing TAG species constituted 79.66% of the total. The profile was dominated by mono-substituted TAGs (57.05%), followed by di-substituted TAGs (20.74%), with only a minor fraction (1.81%) of trisubstituted TAGs. This distribution confirms the efficient and predominantly partial incorporation of NA into the TAG backbone, fundamentally altering the molecular species profile from the base oils and directly influencing subsequent physical properties [30].

2.4.2. Thermal and Crystallization Properties

Thermal properties, critical for application performance, were thoroughly analyzed. As shown in Figure 5A, the SMPs of the NASLs (32.7–33.9 °C) were significantly lower (p < 0.05) than the SMP of the PS base oil (54.4 °C). This reduction is primarily due to the higher unsaturation levels and the integration of long-chain monounsaturated NA, which disrupt the crystalline packing structure [31].

The solid fat content (SFC) profiles (Figure 5B) highlighted the melting behavior of the samples. All the NASLs exhibited a gradual decrease in SFC from 5 to 35 °C, aligning with the desired characteristics of plastic fats. A significant positive correlation was observed between NA content and SFC within the 10 °C to 25 °C range (p < 0.05), indicating that NA enhances the system’s crystallization propensity. The formulation with the highest NA content (CO:PS:NA, 1:1:3) possessed the most pronounced SFC, which confers excellent oil-binding capacity at ambient temperature but may result in excessive hardness when the product is refrigerated, providing information on potential application ranges [32,33].

Differential scanning calorimetry (DSC) analysis (Figure 5C,D) revealed distinct melting and crystallization dynamics. An increased NA proportion transformed the melting endotherm from a bimodal peak to a singular, broader peak, signifying the formation of a homogeneous TAG population through interesterification. This led to a predominant crystal form with a melting range proximate to body temperature, ideal for mouthfeel. Correspondingly, the crystallization exotherms displayed consolidated peaks and higher onset temperatures, suggesting that NA molecules can effectively serve as nucleation templates, fostering a uniform and stable crystal network.

2.4.3. Crystal Polymorphism and Microstructure

The functional performance of plastic fats is largely dictated by crystal morphology. X-ray diffraction (XRD) analysis (Figure 6A) confirmed that all the NASLs predominantly crystallized in the β′ polymorph, as evidenced by the strong short-spacing peaks at 4.15 Å and 3.80 Å, with negligible signals for the β form. The β′ polymorph is highly desirable as its small, needle-like crystals impart a smooth texture, good spreadability, and excellent aeration [34,35].

Polarized-light microscopy (PLM) was used to directly visualize this structure (Figure 6B). The images reveal a network of fine, radially arranged platelet aggregates, typical of β′ crystals. Notably, there is a clear trend of microstructural refinement: as the NA content increased (e.g., from CO:PS:NA= 4:4:9 to 1:1:3), the crystal size became smaller and the distribution more uniform. This refinement can be attributed to NA acting as a co-crystallizing component that modifies crystal growth kinetics, resulting in a denser network of finer crystals. This microstructure directly explains the favorable macroscopic properties, such as enhanced stability and a smooth texture, crucial for high-quality plastic fat applications [36].

Based on the analysis of NASLs predicted through a ML-guided framework, the feasibility of integrating enzymatic esterification with ML has been validated. However, high economic costs remain a key constraint for large-scale application. In this study, we mitigated the expenses associated with pure NA and enzymatic synthesis through process simplification, improved resource utilization, and ML-driven optimization of reaction parameters. Moreover, as a high-value functional ingredient rather than a commodity fat, the product leverages NA’s neuroprotective properties to secure a differentiated niche in markets such as nutraceuticals and functional foods. This positioning partially offsets the costs of raw materials and enzymatic processes. Furthermore, we established an industrialization framework, encompassing ML-optimized operational conditions and a quality-purity correlation model. Future research will focus on cost reduction strategies, including the direct esterification of NA-rich crude extracts (e.g., Acer truncatum and Malania oleifera) with low-cost fats (e.g., palm stearin), thereby providing a dual technical and economic foundation for practical industrial implementation.

3. Materials and Methods

3.1. Materials

Palm stearin (PS), palm kernel oil (PKO), and palm oil (PO) were purchased from Specialty Fats and Oils Technology Co., Ltd. (Shanghai, China). Nervonic acid (NA, 90% purity) was obtained from Shanxi Fuheng Bio-Technology Co., Ltd. (Xi’an, China). Coconut oil (CO) and lard oil (LO) were sourced from laboratory stock. Enzymes, including Lipozyme TL IM (450 IUN/g), Lipozyme RMC (360 IUN/g), and Novozym 435 (9000 PLU/g), were supplied by Beijing Cliscent Technology Co., Ltd. (Beijing, China). Isopropanol (HPLC), n-hexane (HPLC), and other analytical-grade reagents were purchased from Sinopharm Chemical Reagent Co., Ltd. (Shanghai, China).

3.2. Enzymatic Acidolysis

Acidolysis reactions were conducted in a temperature-controlled shaking incubator at 170 rpm. CO, PS, and NA were mixed in a 1:1:1 weight ratio in a sealed flask. Reactions were carried out under varying conditions: temperature (55–70 °C), lipase dosage (6–14% w/w of total substrates), and time (2–14 h). Post-reaction, the enzyme was removed via filtration.

3.3. Separation of Acylglycerols

Acylglycerols were isolated according to a previously reported protocol, with minor modifications [17]. The filtered reaction mixture was dissolved in 10 mL of n-hexane and titrated with 0.5 mol/L of KOH in 30% aqueous ethanol, using 1% phenolphthalein as an indicator. After phase separation, the upper n-hexane layer was collected, dried over anhydrous sodium sulfate, and concentrated under reduced pressure at 40 °C using a rotary evaporator to yield the structured triacylglycerols.

3.4. FA Composition Analysis via GC

Fatty acid methyl esters (FAMEs) were prepared according to an established method, with slight adjustments [31]. Briefly, 20 mg of oil was reacted with 2 mL of 0.5 mol/L NaOH-CH₃OH at 65 °C for 30 min, followed by 2 mL of BF₃-CH₃OH reagent at 70 °C for 10 min. FAMEs were extracted with 2 mL of n-hexane. Analysis was performed using a Shimadzu GC-2030 Plus (Nakagyo-ku, Japan) equipped with a flame ionization detector and a TR-FAME capillary column (60 m × 0.25 mm × 0.25 μm). The oven-temperature program was as follows: 130 °C (3 min), ramp to 200 °C at 5 °C/min (hold for 10 min), and ramp to 220 °C at 2 °C/min (hold for 10 min). The injector and detector temperatures were set to 250 °C. Fatty acids were identified and quantified by comparing retention times and peak areas with known standards according to the method reported by Kim et al. [37].

3.5. TAG Composition Analysis via UPLC-MS

TAG species were analyzed using an AB SCIEX ExionLC UPLC system (Framingham, MA, USA) coupled to a TripleTOF 5600+ mass spectrometer (SCIEX, Framingham, MA, USA). Samples (5 mg/mL in isopropanol) were separated on a Kinetex C18 column (100 mm × 2.1 mm, 2.6 μm) at 40 °C. Mobile phase A consisted of acetonitrile/water/10 mM ammonium formate (60:40, v/v), while mobile phase B was isopropanol/acetonitrile/10 mM ammonium formate (90:10, v/v). The gradient was as follows: 35% B (0–1 min), 35→60% B (1–2 min), 60→100% B (2–13 min), 100% B (13–16.5 min), 100→35% B (16.5–17 min), and 35% B (17–18.5 min) at 0.3 mL/min. MS detection was performed in positive ion mode (m/z 100–1000), with a source temperature of 500 °C and a voltage of 5500 V. TAG quantification was conducted according to the method reported by Fomuso and Akoh with minor modifications [31].

3.6. Additional Analytical Methods

SMP was determined according to the AOCS official method Cc 3b-92 [38]. SFC was measured using a pulsed NMR spectrometer (PQ001–20-010V, NIUMΛG, Suzhou, China) per the AOCS method Cd 16b-93 [39]. Melting and cooling profiles were obtained via differential scanning calorimetry (DSC3, Mettler Toledo, Greifensee, Switzerland) based on AOCS Method Cj1-94 [40]. Crystal structure was analyzed using X-ray diffraction (D2 PHASER, Bruker, Karlsruhe, Germany), and crystal morphology was observed under polarized-light microscopy (PLM, DM2700P, Leica, Wetzlar, Germany).

3.7. Machine Learning Analysis

A dataset comprising fatty acid compositions and corresponding SMP values from 100 enzymatically synthesized structured lipid samples (CO:PS:NA blends) was used to develop predictive models. Fatty acid profiles served as predictor variables (X), while SMP values were used as the response variable (Y). A conventional linear regression model was fitted using OriginPro 2024. Four ML models (gradient boosting, random forest, K-nearest neighbor, and an artificial neural network) were implemented in Python 3.13.2. The optimized hyperparameters were as follows: gradient boosting (10 trees), random forest (5 trees), KNN (K = 10, minimum leaf samples = 3), and an ANN based on the MLPRegressor algorithm [hidden layer sizes = (30, 10, 4), max_iterations = 10,000, activation function = ReLU]. Model performance was evaluated using the coefficient of determination (R²) and root mean squared error (RMSE).

3.8. Data Analysis

All experiments were performed in triplicate, and data are presented as the mean ± standard error of the mean (SEM). We used AB Sciex Cliquid™3.x, MDI Jade 6, Design-Expert 13, and Microsoft Excel for data processing. Statistical analysis and visualization were conducted using OriginPro 2024 and GraphPad Prism 10. A p-value < 0.05 was considered statistically significant. The abbreviations used in this article are listed in Abbreviations.

4. Conclusions

In this study, we present an integrated platform that synergistically combines solvent-free enzymatic interesterification with machine learning (ML) for the rational design of structured lipids (SLs) enriched with nervonic acid (NA). By employing response surface methodology, we optimized the synthesis process, achieving a high NA incorporation rate of 46.76% and a substantial increase in unsaturation (59.34%) under mild conditions (64.6 °C, 7.17 h, 8.46 wt% Lipozyme TL IM). Concurrently, a fatty-acid-composition-driven artificial neural network (ANN) model was established, demonstrating superior accuracy (R² = 0.97) for the sliding melting point (SMP) of NASLs derived from various oil blends, facilitating precise pre-screening of formulations. The resulting SLs exhibited a margarine-compatible SMP range (32.7–33.9 °C), a homogeneous thermal profile, and a stable β′ -form crystal polymorphism with a refined microstructure, collectively confirming their suitability as plastic fats with tailored physicochemical properties. The core contribution of this work lies in our establishment of a closed “synthesis-prediction” loop, offering a generalizable platform for tailoring very-long-chain monounsaturates in next-generation healthy fats.