A Comparative Study of the Sensory and Physicochemical Properties of Cow Milk and Plant‐Based Milk Alternatives

ABSTRACT

There has been an increase in the consumption of plant‐based milk alternatives (PBMA), though these often differ in sensory properties compared to cow milk. Some consumers specifically seek PBMA with sensory qualities that resemble cow milk; however, such products are not yet available on the market. This study compared the chemical composition, particle size, color, viscosity, physical stability, and volatile compounds of seven commercial milk types: almond, coconut, oat, rice, soy, plant‐hybrid (a blend of plant‐based ingredients), and cow milk. Sensory descriptive analysis was used to evaluate appearance, aroma, flavor, and mouthfeel across the samples.

All PBMA, except soy milk, had lower protein levels (<2%) than cow milk (3.2%), as soy naturally contains more protein due to being a legume. PBMA generally had larger particle sizes than cow milk (D_4.3 = 1.41 µm), with rice milk showing the highest (4.19 µm), leading to greater sedimentation and grittiness. Soy milk was linked to beany and astringent attributes, while almond and oat had nutty and cereal notes. Coconut and cow milk were most similar in whiteness (L* > 75) and creamy texture, attributed to their higher fat content and lactones. Plant‐hybrid PBMA, despite lacking cereal ingredients, shared sensory similarities with oat and rice milk.

Volatile compound analysis identified benzaldehyde in almond milk, contributing to its characteristic almond‐like aroma. Coconut milk contained lactones, which are associated with sweet and creamy notes.

Blending complementary PBMA ingredients, such as coconut and soy milk, would effectively mimic cow milk by introducing dairy‐like aroma, creaminess, and improved stability.

Practical Application: There is a need to reformulate PBMA to improve their sensory quality and meet consumer expectations. By understanding how chemical and physical properties influence the overall sensory experience, manufacturers can make better ingredient choices and optimize production processes to produce more desirable products.

1. Introduction

Cow milk is a nutrient‐dense fluid secreted by a female mammal to nourish its offspring (Mehta 2015). It is considered a complete food; hence it is consumed as a source of nutrition by billions of people (Antunes et al. 2023). Despite cow milk being rich in nutrients, some people are unable to consume it for various reasons, including dietary restrictions (such as veganism and vegetarianism), health concerns (such as allergies and lactose intolerance), and environmental considerations such as reducing carbon footprints (Paul et al. 2020; Silva et al. 2022). As a result, the consumption of PBMA made from raw materials like soy, rice, almond, and oat has significantly increased in recent years (Silva et al. 2024). Projections suggest that global revenues of PBMA will rise to USD $38 billion by 2029, up from USD $23 billion in 2023 (Statista 2024).

PBMA are emulsions made from milled plant materials, containing components such as fat, sugars, minerals, and proteins (Mäkinen et al. 2016). They are developed as potential substitutes for cow milk, with some designed to have similar functionality and sensory characteristics (McClements 2020). However, due to variations in production processes and ingredients used, the sensory properties of PBMA, such as taste, color, and mouthfeel, differ from those of cow milk (Sethi et al. 2016; Moss et al. 2023). Some consumers have found these differences to be unacceptable, thus their hesitation to consume PBMA (David J. McClements 2020). For example, oat and soy PBMA have been associated with undesirable beany and earthy off‐flavors, which are due to the presence of lipid oxidation products such as hexanol and hexanal (McCarron et al. 2024; Tangyu et al. 2019). Heat processing has also been found to result in Maillard reaction compounds, including furfurals and pyrazines, in PBMA, which impart a brown color (Manousi and Zachariadis 2019; McClements 2020). PBMA are produced by grinding fibrous raw materials, with the final size determined by the homogenization and milling conditions (Bocker and Silva 2022; Durand et al. 2003), which lead to greater particle size variation compared to cow milk. Larger particles decrease physical stability due to a higher sedimentation rate and can result in an unpleasant mouthfeel of PBMA (McCarron et al. 2024; Maghsoudlou et al. 2016).

Developing PBMA that better mimics cow milk will offer a better option for consumers who prefer and are accustomed to the sensory properties of dairy milk. As a more ecologically sustainable alternative (United Nations 2024), PBMA would potentially drive significant sustainability benefits if widely adopted by the large consumer base of dairy milk. However, no studies have conducted a comparative analysis of cow milk and PBMA, particularly coconut milk and plant‐hybrid formulations, to identify their similarities and differences. To address this comparative knowledge gap, this study aimed to provide insight into the sensory and physicochemical properties of PBMA and UHT full‐fat cow milk. This research will lay the foundation for understanding the ingredients and their application in product development.

2. Materials and Methods

2.1. Materials

A total of five UHT‐treated commercially available milk samples, namely cow milk (full‐fat), oat milk, almond milk, rice milk, and plant‐hybrid milk (prices ranging from AUD $2.00 to 6.00 per/L), were purchased locally (Melbourne, Australia) and stored at room temperature (20‐22°C). The soy milk and coconut milk samples were kindly donated by an industry partner. The sample brands were selected from a pool of market leaders, narrowed down to the final seven, based on the results of an earlier benchmarking exercise conducted by five regular consumers of PBMA and dairy products. These consumers ranked the samples according to sensory characteristics, including taste, aroma, flavor, texture, and appearance (data not shown). Product ingredients are shown in Supplementary Table 1. Reference standards used for sensory descriptive analysis were purchased from local markets (Supplementary Table 2). Reagents (methanol, d₆‐2‐methylpyrazine (99.8 % purity), d₅‐2‐hexanone‐1.1.3.3 (98.9 % purity), d₁₃‐hexanol (98.5 % purity), d₁₂‐hexanal (98.5 % purity), and d₁₁‐3‐methylbutanol (98.9 % purity)) used for headspace—solid phase microextraction gas chromatography‐mass spectrometry (HS‐SPME GC‐MS) analysis were purchased from CDN Isotopes Inc. (Quebec, Canada) and Sigma Aldrich (Victoria, Australia).

2.2. Methods

2.2.1. Sensory Descriptive Analysis

Prior to the commencement of the study, ethics approval was provided by Deakin University (2023‐312) and CSIRO (2021_050_LR W).

A total of ten assessors, comprising nine females and one male between the ages of 18 and 59, were recruited for this study from the pool of trained CSIRO sensory panelists. All assessors had previous experience in descriptive analysis of cow milk and PBMA; therefore, assessors underwent three 2‐h training sessions over three days. During these sessions, assessors generated a list comprising 26 sensory attributes and learned to identify and rate the sensory terms with the assistance of reference standards (Supplementary Table 2). The final training session involved a practice evaluation in individual sensory booths illuminated with 75% white light. Similar conditions were maintained for the formal assessment.

Formal evaluations were conducted over two days, with each session lasting 2 h. Assessors evaluated seven samples per daily session. Each milk type was assessed twice during the formal evaluation. 30 mL milk samples were served in 3‐digit‐coded wine glasses in a randomized order at room temperature (20‐22°C). The assessors rated the appearance, aroma, taste, flavor, and texture/mouthfeel of each sample using a structured continuous intensity line scale numericized 0–100 with indented anchor points of ‘low’ and ‘high’ placed at 5 and 95, respectively. During the evaluation, assessors had access to reference standards and attribute definitions. Assessors were encouraged to use palate cleansers (plain water crackers and water) to limit carryover effects across samples, with a mandatory 2‐min break between samples. They were asked to expectorate samples if desired to minimize the calorie intake. Sensory data was collected using RedJade software (2023, Redwood City, USA).

2.2.2. Particle Size Distribution

The particle size distribution of the samples was determined using a Malvern Mastersizer 3000 (Malvern Instruments Ltd., Worcestershire, UK) according to Jeske et al. (2017) with modifications. Milli‐Q grade water (refractive index 1.33) was used as the dispersing medium for the samples. The refractive index for each milk type was calculated by measuring the Brix value (%) with a handheld pocket refractometer (Atago Co., Ltd., Tokyo, Japan) and converting this value using refractive index tables. The refractive index of the samples ranged from 1.34 to 1.35. Samples were added to the continuously agitated dispersion medium (2500 rpm) until an obscuration rate of 5–6% was achieved. The surface‐weighted mean diameter (D_3,2) and the volume‐weighted mean diameter (D_4,3) were calculated using the Mastersizer 3000 software v 3.88.

2.2.3. Physical Stability

The stability of the milk emulsion was assessed using a Turbiscan Classic 2 (Formulaction, Toulouse, France) according to Matsumiya et al. (2014), with adjustments. Each sample was chilled overnight at 5°C in its original sealed packaging prior to initial analysis to ensure a consistent testing temperature over the measurement period (5 days). After chilling, the package was shaken to ensure uniform mixing, and the sample was then added to a height of 8 cm in a 14 cm borosilicate glass tube with a threaded SVL 15 top and an open bottom. Immediately after sample addition, the tube was positioned in the Turbiscan sample holder for analysis. The Turbiscan measured creaming, sedimentation, and/or flocculation by using light backscattering. Following each measurement, the sample was returned to the refrigerator (5°C) with care to avoid agitation until the next analysis. Measurements were taken every 24 h over the next 4 days and recorded as the percentage of backscattered light as a function of sample height. The sample stability was monitored over 4 days (after day 0), following the recommended usage guidelines of the product.

Turbiscan stability index (TSI) was used to mathematically determine the stability of the milk samples. The TSI was calculated using the formula:

Whereby $n$ is the number of scans; $x_i$ is the backscattering intensity at each height and $x_m$ is the mean backscattering intensity (Zalewska et al. 2019; Sun et al. 2015). The lower the TSI, the more stable the sample (Zalewska et al. 2019)

2.2.4. Viscosity

The viscosity of the milk samples was measured using a digital viscometer (DV‐II+ Version 2.0, Brookfield Engineering Laboratories, Stoughton, USA) equipped with Spindle 00. A volume of 20 mL of each sample was placed into a 30 mL specialized sample holder and equilibrated in a water bath at 25°C for 1 h prior to measurement to ensure temperature consistency.

Viscosity was then measured at a spindle speed of 60 rpm, and readings were initially recorded in centipoise (cP) and subsequently converted to millipascal seconds (mPa·s) (1 cP  =  1 mPa·s). Each sample was measured in four independent biological replicates, with two technical replicates per biological replicate, yielding eight total measurements per sample. Results are reported as mean ± standard deviation. Viscosity values are provided in the supplementary material (Supplementary Table 5).

2.2.5. Color

The color of the PBMA samples was measured using a Minolta colorimeter (Chromameter CR‐300, Konica Minolta Sensing, Tokyo). Calibration was performed using a white plate prior to measuring and in between samples. Samples were filled to a height of 7 cm in transparent, uniform‐walled borosilicate glass tubes (10 cm height × 1.2 cm diameter), which were placed into a sample holder, and color measurements were taken through the side of the tube under controlled lighting and background conditions. For each sample, one reading was recorded, which was the average of readings taken from three different points on the sample. The readings obtained were expressed based on the Hunter values, whereby, a* represented redness (+)/greenness (‐), b* represented yellowness (+)/blueness, (‐) and L* represented lightness from 0 (black) to 100 (white) (McClements 2020).

The degree of whiteness (whiteness index) of the samples was determined using the formula:

$$\begin{array}{ccc}\hfill \mathrm{Whitenes}\mathrm{index}\left(\mathrm{WI}\right)& =& 100-\sqrt{(\left({100-\mathrm{L}\ast )}^2+\mathrm{a}{\ast }^2+\mathrm{b}{\ast }^2\right)}\hfill \\ & & \left(\mathrm{Milovanovic}\mathrm{et}\mathrm{al}.2020\right).\hfill \end{array}$$

2.2.6. HS‐SPME GC‐MS

The volatile aroma compounds in the samples were identified using HS‐SPME GC‐MS according to Nawaz et al. (2021) with modifications. Samples stored at ‐20°C were thawed overnight at ‐5°C. Aliquots (1 mL) were added to a 20 mL headspace vial fitted with 18 mm screw caps (Agilent Technologies, Santa Clara, USA), as well as 0.5 g NaCl and 2 µL internal standard (a mixture of d₆‐2‐methylpyrazine 8.51 mg/L, d₅‐2‐hexanone‐1.1.3.3, 8.74 mg/L; d₁₃‐hexanol, 7.28 mg/L; d₁₂‐hexanal 11.3 mg/L, and d₁₁‐3‐methylbutanol, 14.9 mg/L in methanol). The analysis of the volatile compounds was done using a Shimadzu QP2010 SE model connected to a Shimadzu AOC‐5000 Plus autosampler (Shimadzu Europa GmbH, Duisburg, Germany). The volatiles were extracted using an SPME fiber (divinylbenzene/polydimethylsiloxane/carbon wide range) (Agilent Technologies, Santa Clara, USA) for 30 min at 40°C. The extracted volatiles were desorbed into the GC inlet using splitless mode at 250°C for 1 min using helium as a carrier gas at a 1.56 mL/min flow rate. Thereafter, the volatiles were separated on a DB‐WAX column (length 30 m, diameter 0.25 mm, and thickness 0.25 µm; Agilent Technologies). The oven temperature was then increased to 100°C from 40°C at a rate of 5°C/min. It was then further raised to 250°C at a rate of 10°C/min, where it was held constant for 10 min.

Electron ionization mass spectrometric data were scanned over a range of m/z 35 to 350 with a 0.3 s interval per scan. The ion source temperature was set at 230°C, and the injector temperature at 250°C. Compounds were identified by comparison to the NIST library database (US Department of Commerce, Gaithersburg, MD, USA) and semi‐quantified against the internal standards from similar chemical groups (Supplementary Table 3) using LabSolutions v 4.53SP1 (Shimadzu Europa GmbH, Duisburg, Germany).

2.2.7. Statistical Analyses

Each analysis was carried out using four independent samples, and each sample was measured twice to ensure technical consistency. This resulted in a total of eight replicate measurements per analysis. However, for the sensory evaluation sessions, each milk type was rated in duplicates. The data obtained from instrumental measurements were analyzed and visualized using XLSTAT (v2023.3.1.1416, Lumivero, CO, USA). Mean values were compared using one‐way ANOVA and Tukey's post hoc test at a 5% significance level, and the data was expressed as mean $\pm$ standard deviation. For sensory data, a two‐way ANOVA was conducted using SPSS (IBM SPSS Statistics 29, NY, USA) testing for sample, assessor, and presentation replicate effects and their interactions, treating assessors as a random effect. For principal component analysis (PCA), partial least squares regression (PLSR) and heat maps, data was Pareto scaled and mean‐centered (Næs et al. 2021). For PLSR, the ratings from the sensory descriptive analysis were used as the dependent variables (y) and the physicochemical properties (chemical composition, volatile compounds, physical stability, color, and particle size) were used as the explanatory variables (x). Additionally, the PLSR model was simplified using Variable Importance in Projection (VIP) scores, which allowed the removal of variables with VIP values below the 0.8 threshold, helping to reduce model noise and improve interpretability (Chong and Jun 2005; XLSTAT 2024).

3. Results and Discussion

3.1. Sensory Descriptive Analysis

From the sensory evaluation of milk samples, differences in appearance, aroma, flavor, and texture were found by ANOVA (Supplementary Table 4) and visualized by PCA (Figure 1). Of the 28 sensory attributes rated, 22 terms were found to differ significantly across the samples. No statistical evidence was found supporting differences in sourness, umami, bitterness, saltiness, metallic, and rancid flavor. The largest sensory differences, indicated by the sample effect size (F‐ratio), were coconut aroma, appearance terms, dairy aroma, and gritty texture (Supplementary Table 4).

FIGURE 1.

Principal component analysis on almond, coconut, oat, rice, soy, plant‐hybrid, and cow milk types and sensory attributes.

PC1 accounted for 34.49% of the variance, and PC2 22.98%. Examination of the plots revealed four distinct sample groups: almond milk, rice/plant‐hybrid/oat milk, soy milk, and cow/coconut milk. PC1 mainly discriminated between soy samples (right side of PC1) and cow and coconut milk (left side of the axis). PC2 primarily distinguished almond milk (left side of the axis), from rice, plant‐hybrid, and oat milk (right side). The cereal‐based PBMA (rice and oat milk) exhibited a degree of similarity, as they were characterized by the same attributes, such as cereal and caramel notes. Although not supported by statistical evidence, other characteristics associated with these two PBMA samples included vanilla flavor non‐significant (ns), bitterness (ns), saltiness (ns), sourness (ns), and brown color. This is in agreement with the findings of Pointke et al. (2022) and Vaikma et al. (2021), who also reported that assessors associated oat PBMA with bitterness, cereal notes, saltiness, and brownness. The plant‐hybrid milk did not have cereal as its primary ingredient but was still classified similarly to cereal‐based PBMA. This similarity may be attributed to the presence of natural flavors (Supplementary Table 1), which could have contributed to a cereal‐like aroma. Another possible explanation is surface mimicry, where the visual similarity between the plant‐hybrid milk and cereal‐based PBMA (brown color) may have led assessors to transfer attributes, such as cereal aroma, from rice and oat PBMA (Van Kerckhove et al. 2022).

The similar characterization of cow and coconut milk may have been due to the influence of visual cues on the assessors. Assessors scored the intensity of white color as high in both samples, which may have caused them to perceive comparable coconut and dairy notes in both coconut and cow milk, as also observed by (Wadhera and Capaldi‐Phillips 2014). As was expected, the assessors also associated viscosity and dairy notes with creaminess in the same samples (coconut and cow milk). This was due to creaminess being a complex attribute that involves multiple factors, including smoothness, mouth‐coating, dairy aroma, and thickness (Flett et al. 2010; Richardson‐Harman et al. 2000).

Other correlations were also displayed by the PCA, which have been reported prior in literature. These included the nutty and chocolate attributes of almond milk and the astringent, metallic, beany, umami, rancid, and yellow color of soy milk (Vaikma et al. 2021; Gorman et al. 2021). However, contrary to the previous studies, the soy milk in this study was also associated with sweetness. The sweetness score in soy milk was likely due to the addition of sugar to the PBMA in order to improve palatability and mask bitterness and astringency (Eric Walters 1996; Pointke et al. 2022). However, soy milk was still rated highly in astringency, suggesting ineffective masking by the added sugars (Pointke et al. 2022).

3.2. Chemical Composition

The proximate composition listed on each product's chemical composition (Table 1) showed that both cow milk and PBMA contain common major components like protein, carbohydrates, sodium, dietary fiber, and fat, though in varying amounts. These differences were attributed to the various plant sources and production processes, including formulations and extraction methods (Kyriakopoulou et al. 2021; Mäkinen et al. 2016).

TABLE 1.

Chemical composition of almond, coconut, oat, rice, soy, plant‐hybrid, and cow milk types per 100 g.

Milk type	Carbohydrates (g)	Protein (g)	Sugar (g)	Dietary fiber (g)	Total fat (g)	Sodium (mg)
Almond milk	2.8	0.8	1.7	N/A	2.5	41
Coconut milk	3.0	0.6	2.6	N/A	3.7	48
Oat milk	6.1	0.8	1.8	0.4	3.0	48
Rice milk	10.3	0.6	3.6	<1	1.2	68
Soy milk	5.5	4.1	2.2	1.4	2.2	47
Plant‐hybrid	1.8	1.6	1.2	N/A	3.3	62
Cow milk	4.8	3.2	4.8	N/A	3.4	45

Note: Carbohydrates include sugars. Values obtained from nutrition information panel on product packaging.

Major differences between cow milk and PBMA were in primary ingredients and their respective concentrations (oats ‐ 9%, soy ‐ 14%, almond ‐ 3.5%, rice ‐ 14%, coconut ‐ 15% and plant‐hybrid—undisclosed) (Supplementary Table 1). Such differences in primary ingredients and additives (oil type, thickeners, and emulsifies) would affect the resultant proximal composition of the product, and in turn, their physiochemical and sensory properties (Daszkiewicz et al. 2024). A notable difference was the type of sugar present; cow milk contained lactose, a disaccharide (Table 1), while some PBMA contained sucrose (Supplementary Table 1). This difference could potentially affect the sweetness intensity in the products, as lactose is one‐sixth as sweet as sucrose (Pang and Zhang 2021). Furthermore, during PBMA production, carbohydrates such as cellulose and starch are hydrolyzed (McClements 2020), breaking them down into sugars (fructose and glucose) that are more reactive than lactose (Lund and Ray 2017). Therefore, there would be a higher rate of heat‐induced compound formation in PBMA compared to cow milk. These differences could further accentuate sensory differences between PBMA and cow milk. Another difference lies in the protein content. Most PBMA had low protein levels (less than 2%), except for soy milk, which contained notably higher amounts. This is likely because soy is a legume, which naturally has a higher protein content, whereas other plant‐based sources tend to be lower unless the product is fortified (Pointke et al. 2022).

3.3. Particle Size Distribution and Physical Stability

Particle size plays a major role in the physical stability of milk, as it is one of the factors that determine the rate at which the dispersed phase (droplets) separates from the continuous phase (water) (Durand et al. 2003). The D₅₀ diameters for all milk types ranged from 0.84 ‐ 1.34 µm, while the D₉₀ diameters exhibited an even wider range of particle sizes, spanning from 1.50 ‐ 8.16 µm, indicating a lack of uniformity (Table 2). Similar to cow milk and the findings of Amador‐Espejo et al. (2014), the PBMA exhibited a bimodal distribution, with the first peak significantly higher than the second, and a tail end, signifying the presence of a few larger particles (Figure 2). The larger particles (second peak and tail end) may have formed due to droplet‐droplet collisions that occur post homogenization, resulting in coalescence (McClements et al. 2022).

TABLE 2.

Particle size characterization of almond, coconut, oat, rice, soy, plant‐hybrid, and cow milk types.

Milk type	D50 (µm)	D90(µm)	D3,2(µm)	D4,3(µm)
Almond	1.31 ± 0.01 b	3.20 ± 0.13bc	1.24 ± 0.01b	1.94 ± 0.02b
Coconut	1.34 ± 0.01 a	3.59 ± 0.12b	1.27 ± 0.05a	1.88 ± 0.03c
Oat	0.88 ± 0.04 e	1.60 ± 0.02d	0.82 ± 0.05e	1.03 ± 0.01e
Rice	1.26 ± 0.01c	8.16 ± 0.38a	1.21 ± 0.01c	4.19 ± 0.04a
Soy	1.01 ± 0.03d	3.02 ± 0.66c	0.96 ± 0.03d	1.89± 0.02c
Plant‐hybrid	0.84 ± 0.01f	1.50 ± 0.06d	0.78 ± 0.01f	1.01 ± 0.03e
Cow	0.87 ± 0.02e	1.74 ± 0.16d	0.83 ± 0.02e	1.41 ± 0.02d

Note: Values presented as mean values of 4 repetitions (n = 4) ± standard deviation. Values in the same column with different superscript letters differ significantly (p < 0.05). D₅₀ represents the diameter below which 50% of the volume of particles are found; D₉₀ represents the diameter below which 90% of the volume of particles are found; D_3,2 (Sauter diameter) represents the surface‐weighted mean diameter; D_4,3 represents the volume‐weighted mean diameter.

FIGURE 2.

Particle size distribution of almond, coconut, oat, rice, soy, plant‐hybrid, and cow milk types. Each milk type is presented as mean values of four repetitions  (n= 4) ± standard deviation.

D_4.3 and D_3.2 parameters were used to characterize the larger and smaller particles present in the emulsion, respectively (Thiebaud et al. 2003). There was no statistical difference observed between the particle size of cow milk, plant‐hybrid milk, and oat milk. This similarity suggests that the oat and plant‐hybrid milk utilized efficient particle reduction processes. PBMA contains milled fibrous plant matter and other ingredients that can influence particle size negatively (Durand et al. 2003; Antunes et al. 2023). Hence, during production, PBMA slurry undergoes hydrolysis, allowing improved solubility of plant carbohydrates (McClements 2020) to help reduce particle sizes. However, oat milk typically undergoes an additional decanting step, where insoluble material such as fiber, bound proteins, and minerals are removed (Zhang et al. 2007), potentially leading to more refined particle sizes similar to homogenized cow milk.

Changes in ingredient choices may have also influenced particle size distribution. Unlike all other milk types, the plant‐hybrid milk contained pea protein (Supplementary Table 1), which possesses the ability to partially unfold and adsorb to the fat droplet, forming a viscoelastic layer (Ozturk and McClements 2016). This layer may have prevented droplet coalescence, thereby maintaining smaller particle sizes post‐production. Moreover, a more efficient homogenization process may have also been adopted. It could be that the smaller‐sized samples (cow, plant‐hybrid, and oat) were subjected to a higher homogenization pressure to effectively reduce particle sizes or a two‐stage homogenization process to prevent droplet aggregation (Thiebaud et al. 2003; Bernat et al. 2015). Other processes, including ultra‐high thermal processing, also affect particle size distribution by increasing the hydrophobicity of the non‐polar amino acids and interactions between proteins, ultimately leading to coalescing and sedimentation (Mäkinen et al. 2016). Apart from influencing sample stability, sample particle size determines the degree of grittiness in a sample. Larger particles than those of cow milk would affect the mouthfeel and texture, potentially leading to an undesirable gritty texture.

Physical stability was assessed using a Turbiscan, which monitored the backscattered light intensity through the sample over time. The Turbiscan graph (Figure 3) showed that on day zero, each sample exhibited consistent backscattering levels, indicating a uniform distribution of particles in the emulsions. These results were anticipated, given the thorough mixing conducted on the samples before the initial measurement. By days 2 and 4, none of the samples exhibited this consistent pattern (Figure 3), suggesting a shift in particle distribution and a departure from a homogeneous mixture. Emulsions are thermodynamically unstable and therefore are prone to separate due to instability mechanisms that include creaming, gravity, and aggregation (Piorkowski and McClements 2014). For cow milk, an increase in the backscattering on the upper half of the sample (40 ‐ 80 mm) was observed as well as a decrease in the bottom half (below 40 mm). This observation suggested that cow milk had begun creaming, a process where the fat globules migrate to the top of the emulsion due to lower density (Meng et al. 2022). This migration results in a higher fat concentration at the top and consequently greater backscattering (Durand et al. 2003). Similarly, all the other PBMA, apart from rice milk, showed an increase in backscattering above 40 mm, which could also be attributed to the upward movement of fat globules. Rice milk contained very low concentrations of fat (1.6%) (Table 1); hence the fat globules may have had very little scattering effect despite possibly migrating to the top of the sample. Additionally, rice milk contained the largest particles (D_4.3 = 4.19 µm) (Table 2), causing them to settle at the bottom faster than those of the other samples, as was also observed by Mäkinen et al. (2015). Some of these settled particles (plant material) may have had scattering abilities (McClements 2020), which could explain the observed increase in backscattering on days 2 and 4 at the bottom (below 50 mm) of the rice milk tube. The oat and plant‐hybrid milk also showed a high degree of backscattering at the bottom of the tube (< 10 mm), also likely due to the settling of plant material with light‐scattering abilities.

FIGURE 3.

Changes in backscattered light intensity according to height, in rice, soy, plant‐hybrid, and cow milk samples on (a) day 0, (b) day 2 and (c) day 4. Each sample is presented as the mean value of four repetitions (n = 4) 3a: Backscattering intensity versus sample height of almond, coconut, oat, rice, soy, plant‐hybrid, and cow milk types on day 0. Each milk type is presented as mean values of four repetitions  (n= 4) ± standard deviation. 3b: Backscattering intensity versus sample height of almond, coconut, oat, rice, soy, plant‐hybrid, and cow milk types on day 2. Each milk type is presented as mean values of four repetitions  (n= 4) ± standard deviation. 3c: Backscattering intensity versus sample height of almond, coconut, oat, rice, soy, plant‐hybrid, and cow milk types on day 4. Each milk type is presented as mean values of four repetitions  (n= 4) ± standard deviation.

While not significantly different from the oat and plant‐hybrid milk, the coconut and almond milk had the lowest TSI (1.47 and 1.85, respectively) in the sample set (Figure 4). As previously mentioned, particle size affects the stability of PBMA; however, the coconut and almond milk did not possess the smallest particles (Table 2). Therefore, the higher stability may be attributed to a greater protein solubility of their proteins specific to their formulations (coconut and almond), resulting in a lower rate of precipitation. This is in agreement with the findings of Roland et al. (2024), who found that PBMA with more soluble proteins was the most stable despite a larger particle size.

FIGURE 4.

Turbiscan Index Stability (TSI) of almond, coconut, oat, rice, soy, plant‐hybrid and cow milk types. Lower TSI indicates increased stability. Superscript letters indicate significant differences (p < 0.05). Rice milk was excluded as it was an outlier affecting statistical analysis.

Emulsification in PBMA functions similarly to that in cow milk, where casein micelles from within the serum adsorb onto the surface of the newly formed fat droplets to create a stable emulsion by steric repulsion (Dickinson 1999). However, cow milk was not the most stable product, despite containing casein and relatively smaller particle sizes, suggesting that additional stabilizers used in PBMA contributed to improved stability. Viscosity may have also played a key role in the higher stability of coconut and almond milk, as they were the most viscous samples (Supplementary Table 5). This viscosity was due to the use of thickeners (vegetable gums), which form a gel network (Himashree et al. 2022), restricting particle movement and slowing their migration to the top or bottom. Similar observations have also been reported by Meng et al. (2023). However, using thickeners in PBMA may result in a more viscous texture, creating a notable sensory contrast to cow milk that may be perceived as a vast difference. Ensuring good emulsion stability in PBMA throughout its shelf life is crucial, as it remains a significant challenge in the industry (Jeske et al. 2019).

3.4. Color

3.4.1. L*, a*, and b* values

Colorimetry was used to characterize the L*a*b* color space of all samples. The L* value (lightness) was determined by the degree of light scattering caused by particles in the milk, including fat and proteins (Katsumata et al. 2020). Cow milk did not contain the highest protein (3.2% compared to 4.1% in soy milk) or fat (3.4% compared to 3.7% in the coconut milk) content amongst the samples, yet it exhibited the highest L* value (77.58) (Table 3). According to the findings of Misawa et al. (2016) and Cheng et al. (2019), the light scattering effects of protein and fat are not additive; hence it could be concluded that the high L* value in cow milk was likely due to the presence of casein micelles, which were present only in cow milk. Coconut milk was the second lightest sample due to its high fat concentration (3.7%, Table 1). Rice, soy, and the plant‐hybrid milk had fat concentrations of 1.2%, 2.2%, and 3.3%, respectively, with protein contents of 0.6%, 4.1%, and 1.6% (Tables 1). Despite these differences, their lightness values were statistically similar (Table 3). This may be attributed to the presence of calcium carbonate, a white colorant, which was included in all three PBMA formulations (Supplementary Table 1) to achieve similar lightness (Younes et al. 2023). Oat milk had a higher L* value than almond milk, likely due to a combination of its higher fat content and smaller particle size compared to almond milk (2.5% fat, D_4.3 = 1.94 µm) (Tables 1 and 2, respectively). The higher fat concentration and smaller particles could have had an additive effect on the extent of light scattering, thus increasing the L* value (McCarron et al. 2024; Misawa et al. 2016).

TABLE 3.

Effect of almond, coconut, oat, rice, soy, plant‐hybrid and cow milk types on color.

Milk type	L*	a*	b*	Whiteness index
Almond	69.58 ± 0.52d	−0.01 ± 0.04a	4.72 ± 0.06c	69.22 ± 0.51e
Coconut	74.30 ± 0.53b	−1.39 ± 0.02c	2.22 ± 0.09d	74.17 ± 0.52b
Oat	72.23 ± 1.04c	−0.01 ± 0.04a	8.07 ± 0.16a	71.07 ± 0.96d
Rice	72.78 ± 0.68bc	−0.73 ± 0.06b	−1.54 ± 0.61e	72.72 ± 0.66bc
Soy	72.82 ± 0.68bc	−1.70 ± 0.08d	7.69 ± 0.21a	71.70 ± 0.59cd
Plant‐hybrid	72.78 ± 0.68bc	−0.73 ± 0.06b	5.66 ± 0.13b	72.19 ± 0.64cd
Cow	77.58 ± 0.52a	−2.55 ± 0.06e	4.84 ± 0.14c	76.92 ± 0.48a

Note: Values presented as mean values of 4 repetitions (n = 4) ± standard deviation. Values in the same column with different superscript letters differ significantly (p < 0.05). Hunter values represented as (L*, a* and b*). (L*) represents the lightness scale, (a*) represents the green and red scale (indicated by negative and positive signs, respectively), and (b*) represents the blue and yellow scale (indicated by negative and positive signs, respectively).

The a* value indicates the intensity of redness or greenness in a sample and is determined by the wavelength of light absorbed by the sample particles (Cheng et al. 2019). Hence, if a particle absorbs the red light, the green color is reflected and observed in the milk sample. The a* values of all the samples were consistently negative and thus associated with greenness (Table 3). Cow milk had the highest greenness index, followed by soy, coconut, rice, and plant‐hybrid oat and almond milk types, respectively. In cow milk, the green color is associated with the presence of riboflavin (vitamin B2) (Misawa et al. 2016); therefore, it could be that the green color in PBMA was due to pigments such as chlorophyll (McClements 2020). However, the a* values were all close to zero, relative to the scale range of ± 128. Hence, the level of greenness was subtle, as a value of zero on the scale is consistent with a neutral grey color (Scarso et al. 2017).

A similar explanation could be given for the b* values, where most of the milk types, except for rice milk, were yellow (Table 3). The yellowness in cow and soy milk was likely due to the presence of β‐carotene in the samples. In the case of cow milk, this pigment is absorbed from the animals' diet, while in soy milk, this pigment is naturally present in soybeans (Conboy Stephenson et al. 2021; Iwuoha and Umunnakwe 1997). Another potential explanation for the yellowness observed in the milk samples could be attributed to the presence of Maillard reaction products such as melanoidin pigments formed during heat treatment of the milk (Morales and van Boekel 1998; Rauh and Xiao 2022).

3.4.2. Whiteness Index

According to Milovanovic et al. (2020), the WI mathematically combines the L*, a*, and b* values to give a single term ranging from 1 to 100 that can be used to assess the color of milk. The WI of each milk type examined in this study fell within a comparable range of 69.22 ‐ 76.92 (Table 3), aligning closely with those reported by McClements (2020). From these values, it could be deduced that none of the samples achieved absolute whiteness. However, they did exhibit a greater degree of whiteness than darkness, given that all index values exceeded 50. Compared to cow milk (76.92), the whiteness index of all PBMA was significantly lower, in particular that of almond milk (69.22), consistent with the findings of Jeske et al. (2017). Coconut milk (74.17), however, was closest to cow milk, and this colorimetric similarity was reflected in the white appearance ratings from the sensory evaluation. Color can influence perceived taste; for example, consumers may associate yellowness with sourness or pink with sweetness (Woods and Spence 2016; Yang et al. 2024). Thus, deviations in PBMA from the whiteness of cow milk (appearing darker, greener, or redder) could further detract from its resemblance to cow milk, underscoring the importance of careful ingredient selection.

3.5. Volatile Aroma Compounds

A total of 31 volatile compounds were identified across all samples, comprising seven alcohols, 11 aldehydes, three ketones, four pyrazines, one furan, four lactones, and one terpene (Supplementary Table 3 and Figure 5). Certain compounds, such as benzaldehyde, hexanal, and 1‐pentanol, were present across all samples, however in varying quantities.

FIGURE 5.

Heatmap illustrating the relative concentrations of volatile compounds in almond, coconut, oat, rice, soy, plant‐hybrid, and cow milk types. Each rectangle in the heatmap represents the relative concentration of a single compound in each milk type. The concentrations are indicated by a color gradient from blue to red through white, representing increasing concentrations.

Almond milk contained one of the largest amounts of benzaldehyde, aligning with the findings of Pointke et al. (2022), who reported high concentrations of benzaldehyde in almond milk, a key aroma compound to almond aroma. Benzaldehyde, as previously noted, was also present in all other milk types. However, unlike in almond milk, it was not a characteristic compound but rather a result of processing conditions, particularly the Maillard reactions induced by heat treatment (Jo et al. 2018).

Coconut milk was unique in containing a variety of lactones (γ‐decalactone, δ‐octalactone, ϒ‐octalactone, and δ‐undecalactone), as no other type of milk contained these compounds. This unique compositional difference likely explained why coconut aroma strongly differentiated the samples in the sensory evaluation. According to (Wang et al. 2020), lactones constitute 34% of volatile compounds in coconut milk, making them the most abundant compounds in the milk. Lactones are derived from fatty acids and have been associated with the nutty, creamy, and sweet odors of coconut milk (Vaikma et al. 2021; Tinchan et al. 2015). Aliphatic lactones such as (Z)‐6‐dodecen‐4‐olide, which were not measured here, are also common to dairy products and could explain why coconut and cow milk were rated similarly for some aroma attributes (Figure 1) (Dimick et al. 1969).

Similar to the findings of McCarron et al. (2024), hexanal was the most abundant aldehyde in oat milk. While hexanal was a common compound across the sample set, its concentrations varied significantly, with the plant‐hybrid milk exhibiting the highest levels. Hexanal is formed through lipid oxidation, particularly during storage, and is associated with PBMA off‐flavors, such as bean‐like and earthy notes (Pointke et al. 2022; van Aardt et al. 2005).

The major compounds in rice milk were aldehydes (pentanal, benzaldehyde, and nonanal) and alcohols (1‐pentanol and 1‐octen‐3‐ol). Some of these compounds, such as 1‐pentanol (fruity odor), nonanal (citrus odor), and 1‐octen‐3‐ol (mushroom odor), are known to contribute to the aroma of cooked rice (Yang et al. 2008; Friedrich and Acree (1998)).

Several methyl ketones were commonly found in different samples, including 2‐heptanone (in cow, almond, and plant‐hybrid milk), 2‐nonanone (in almond, coconut, rice, and plant‐hybrid milk), and 2‐octanone (in coconut, oat, rice, soy, and plant‐hybrid milk). Cow, almond, and plant‐hybrid milk also contained pyrazines, 2,6‐dimethyl pyrazine and 2,5‐dimethyl pyrazine. Furan compounds were also present, with 2‐pentylfuran found in coconut, oat, rice, soy, and plant‐hybrid milk. The formation of pyrazines compounds was due to the Strecker reaction, while the furans, aldehydes including benzaldehyde, and methyl ketones are Maillard reaction products (Xi et al. 2023; Manousi and Zachariadis 2019; Jo et al. 2018).

Alcohols were also detected in the samples, with the most common being 1‐pentanol (coconut, oat, rice, soy, plant‐hybrid, and cow milk), 1‐octen‐3‐ol (coconut, rice, soy, almond, and plant‐hybrid milk), 1‐octanol (coconut, rice, plant‐hybrid, almond, and cow milk), 1‐heptanol (rice, plant‐hybrid, and cow), and 3‐methyl‐1‐butanol (rice, almond, and cow milk). These volatiles have also been reported by Pointke et al. (2022) and Manousi and Zachariadis (2019).

Several studies have demonstrated that the rate of chemical reactions forming the compounds discussed above is affected by various factors during production, as well as by the initial chemical composition and storage conditions, leading to variations in compound concentrations. Roasting almonds in the production of almond PBMA has been shown to increase the concentrations of pyrazines and alcohols, while a high initial unsaturated fat concentration may provide fatty acids for the formation of ketones via β‐oxidation (Xiao et al. 2014; Pérez‐González et al. 2015). Additionally, it has been reported that storage conditions such as temperature and exposure to light can induce lipid oxidation, leading to the creation of other compounds, including 2‐heptanone (van Aardt et al. 2005; Xi et al. 2023). The formation of these compounds would contribute to the characteristic aromas of these products as well as the formation of off‐flavors and color changes during production and storage.

GC‐MS analysis showed that cow milk contained similar compounds to other PBMA. For instance, both coconut milk and rice milk contained 2‐octanol, while almond milk also had 2‐heptanone, along with Maillard reaction products such as 2,5‐dimethylpyrazine and benzaldehyde. Additionally, almond milk contained furfuryl alcohol as well as octanal, which was found in all samples except soy milk. Furthermore, 3‐methylbutanal and 2‐octanone were present in all milk types. Markedly, coconut milk contained lactones, which are important for imparting the creamy, dairy‐like aroma. The presence of these similar compounds, along with compounds that give off similar flavors and aromas, suggests that blending different milk types and manipulating the concentrations of these compounds could enable the development of PBMA that closely replicates the aroma and flavor of cow milk.

3.6. Relationship Between Physicochemical Properties and Sensory Attributes

The relationship between the physicochemical properties of the milk samples (x‐variables) and their significant sensory attributes (y‐variables) was further analyzed using partial least squares regression (PLS‐R). The importance of the x‐variables was determined by jackknifing and the size of standardized regression coefficients (β) to evaluate the relative strength and direction of their effects on y‐variables, also considering the goodness of fit indicated by the R2 value for each sensory term. The cumulative indices for R²Y and R²X were 78% and 82%, respectively, indicating that the model explained a substantial portion of the variation.

The PLSR model (Figure 6) showed that some x‐ and y‐variables were positioned between the inner and outer ellipses, suggesting they were well‐modeled (R² = 0.5 to 1). For instance, the white color was effectively modeled (R² = 0.92), with positive regression coefficients indicating a positive association with total fat (β = 0.09) and whiteness index (WI) (β = 0.06). This observation corroborates the earlier discussion in sections 3.1 and 3.4, that the higher whiteness in coconut milk, as noted by both sensory assessors and instrumental measurements, is primarily due to its fat content, which enhances whiteness and increases the WI (Table 3).

FIGURE 6.

Partial least squares regression (PSLR) plot of correlation between physicochemical properties and sensory attributes of almond, coconut, cow, plant‐hybrid, soy, oat, and rice milk types. R²Y cum = 0.78 and R²X cum = 0.82.

The nutty flavor (R² = 0.95) was associated with benzaldehyde (β = 0.07), 1‐heptanol (β = 0.07), 2,6‐dimethyl pyrazine (β = 0.07), and 2,5‐dimethyl pyrazine (β = 0.06). These compounds have been previously linked to nutty flavors in various studies (Pointke et al. 2022; Vaikma et al. 2021). Astringency (R² = 0.87) was weakly associated with protein (β = 0.03) but more related to dietary fiber (β = 0.061). Dietary fiber may contain polyphenols, which, along with plant proteins, are known astringents that reduce oral friction (Pires et al. 2020). Beaniness (R² = 0.82) was strongly positively associated with 2‐heptanal(E) (β = 0.13), a compound linked to green aromas (McCarron et al. 2024), which may affect consumer acceptance of PBMA.

Dairy‐like sensory attributes such as mouthfeel and texture are key factors in PBMA acceptance for consumers who prefer a sensory experience similar to cow milk (McClements 2020). However, replicating its characteristic creaminess remains challenging, as PBMA often exhibit distinct textural properties. In this study, Grittiness was effectively modeled (R² = 0.96) and was strongly associated with increased particle size (β = 0.11) and TSI (β = 0.11), while it showed a negative relationship with total fat (β = ‐0.09). Conversely, creaminess—also well modeled (R² = 0.99)—was primarily linked to total fat (β = 0.11) and a series of lactones (β = 0.06), while being negatively associated with sodium (β = ‐0.10), TSI (β = ‐0.10), particle size (β = ‐0.09), and 1‐octen‐3‐ol (β = ‐0.11). The latter compound, which may contribute to a mushroom aroma (Yang et al. 2008), could have been incongruent with creaminess, diminishing its perceived intensity through cross‐modal interactions.

Based on PLS‐R analysis, the physicochemical properties of milk significantly influenced its sensory attributes, with variations largely driven by raw materials and chemical composition. This underscores a key distinction between cow milk and PBMA, as their differing compositions ultimately shape flavor and texture. Key findings included the impact of fat on whiteness and creaminess and associations between specific compounds and nutty, astringent, or beany notes. These results highlight the intricate relationship between composition and sensory properties in milk, emphasizing the unique characteristics of cow milk versus PBMA. Future PBMA formulations aiming to replicate the sensory properties of cow milk could increase fat content, reduce particle size, and remove or mask incongruent plant‐derived flavors. Additionally, dairy‐like flavors, such as creaminess, can be achieved by adding coconut milk, which contains lactone compounds that impart these characteristic notes. Additionally, blending PBMA ingredients like soy milk, which provides high protein to enhance emulsification and stability, with coconut milk, known for its dairy‐like aroma, whiteness, and creaminess, could be an effective strategy for product development. The high fat content in coconut milk could also aid in masking astringency (Pires et al. 2020), further enhancing the product's similarity to cow milk.

4. Conclusions

This study highlighted how physicochemical properties such as particle size, viscosity, color, and volatile compounds influence the sensory characteristics of cow milk and PBMA. Cow milk was characterized by higher whiteness and consistent creaminess, though some PBMA, such as coconut and almond milk, also exhibited good stability and creamy texture. Overall, PBMA showed greater variation in flavor and physical stability depending on plant source and formulation.

While no single PBMA matched cow milk across all sensory attributes, several shared overlapping traits such as creaminess, whiteness, or aroma. Interestingly, despite the absence of cereal ingredients, the plant‐hybrid PBMA displayed cereal‐like notes similar to oat and rice milk, indicating that blending complementary plant‐based ingredients, such as coconut and soy, can impart desirable characteristics like those found in dairy milk. Also, other milk types, such as almond milk, whose aroma compounds can enhance the fatty, burnt, and nutty attributes typically associated with UHT milk. Furthermore, the similarity of PBMA to cow milk may be improved by reducing particle size, increasing fat content, and introducing masking agents.

While consumer acceptance was not evaluated, the findings offer a foundation for developing PBMA formulations that more closely resemble cow milk in sensory and functional characteristics. Future research should explore consumer acceptance of these optimized PBMA formulations to determine how closely sensory similarity aligns with consumer preference.

Author Contributions

Anesu Avril Magwere: conceptualization, investigation, writing–original draft, methodology, visualization, writing–review and editing, formal analysis, data curation, validation. Russell Keast: conceptualization, supervision, writing–review and editing, methodology. Shirani Gamlath: conceptualization, writing–review and editing, methodology, supervision. Damian Espinase Nandorfy: writing–review and editing, methodology, formal analysis. Nelum Pematilleke: Methodology, writing–review and editing. Joanna M. Gambetta: conceptualization, methodology, writing–review and editing, supervision.

Conflicts of Interest

The authors declare no conflicts of interests.

Supporting information

Supplementary Tables: jfds70370‐sup‐0001‐SuppMat.docx