Research on donkey milk · Study 113

Milk miRNA on the computer: donkey, human, cow, goat

From 29 public sequencing data sets, including 4 from donkey milk, 1549 conserved miRNAs were recorded. miR-148a-3p made up the largest share in all four types of milk, on average about 21.5 %.

Important: A laboratory analysis of milk samples. It describes properties of the milk and does not examine any effect of donkey milk in humans.

At a glance · swipe or use the arrows

The key figures

29public data sets (human 8, cow 13, goat 4, donkey 4)

1549conserved miRNAs across all samples

456of which detected in all four types of milk

≈ 21.5 %mean share of miR-148a-3p in the total miRNA signal

≈ 73.7 %share of the 20 most frequent miRNAs

4577genes in the analysis after noise filtering (6039 excluded)

1809genes in the shared core of all four types of milk (39.5 %)

770genes with a predominantly donkey-related score (16.8 %)

770 vs. 325 ± 16donkey category: observed vs. expected under random distribution (permutation test)

0.67rank correlation of scores, donkey vs. human (cow vs. goat: 0.93)

0.82mean silhouette width of grouping by species

6.68 %share of mappable reads in one of the four donkey samples

Study design

How it was studied

A purely computational analysis without the authors’ own samples: publicly deposited sequencing data on small RNAs from 29 milk or milk cell samples from human (n = 8), cow (n = 13), goat (n = 4) and donkey (n = 4) were reanalysed uniformly. The four donkey samples come from two published comparative studies on milk vesicles from livestock. The miRNAs were quantified against human reference sequences, so only miRNAs conserved between species were captured. For each human gene, a score weighted by miRNA abundance was calculated per type of milk from predicted and experimentally supported miRNA–gene pairs from two public databases; genes were divided into 16 categories according to the ratio of the scores. The analysis included enrichment analyses, permutation and sensitivity tests and cluster analyses of the individual samples. The authors take uptake of the miRNAs via food as a purely working assumption.

Laboratory study

Study typeLaboratory studyCountryItalyYear2026JournalInternational Journal of Molecular Sciences

Interactive chart · values from the study

Largest gene categories in the target gene profile

Table 1 of the study; the six smallest categories (224 genes together) are not shown. Computational predictions, not measurements in cells or humans.

Second analysis

Cumulative share of the most frequent miRNAs

Rounded values from section 2.1 and Figure 1A; top 1 is miR-148a-3p.

Results

What the study measured

Across all 29 data sets, 1549 conserved miRNAs were found, 456 of them in all four types of milk. The spectrum was strongly concentrated on a few miRNAs: miR-148a-3p was the most frequent miRNA in every type of milk (on average about 21.5 %), the five most frequent made up about 42.5 % and the 20 most frequent about 73.7 %. The composition of the 15 most frequent miRNAs was similar between species and differed mainly in quantity.

After noise filtering, 4577 predicted target genes remained. 1809 of them (39.5 %) formed a shared core with similarly high scores in all four types of milk; with 770 genes (16.8 %), the donkey-related category was the second largest, followed by the human-related one with 430 genes.

In the permutation test, the size of the donkey category, at 770 genes, was well above the random expectation of 325 ± 16 (p < 0.001). The scores of cow and goat correlated most strongly (ρ = 0.93), those of donkey and human most weakly (ρ = 0.67).

In principal component and cluster analysis, the 29 individual samples grouped by species (mean silhouette width 0.82); donkey milk was the most clearly separated from the other types of milk. The category structure remained stable at thresholds between 1.3 and 2.0.

Data quality was uneven: in the samples from one source project, 59.5–76.6 % of reads could be mapped to human references, but in the donkey sample from the second project only 6.68 %. According to the authors, the donkey spectrum remained practically unchanged without this sample (ρ = 0.99).

The authors explicitly described the enrichment analyses as hypothesis-generating. For genes in the shared core, they noted a methodological bias: these are targeted on average by 15.3 rather than 8.1 miRNAs, which favours enrichments.

Authors’ conclusion: In the view of Zoziuk et al. (2026), the predicted target gene profile of miRNAs from human, cow, goat and donkey milk is based on a large shared core shaped by a few frequent miRNAs such as miR-148a and the let-7 family; on top of this are species-specific patterns, among which the donkey-related group is the second largest. The authors stressed that these are computational predictions under the assumption of uptake via food that do not constitute experimental confirmation, and regarded their data as a basis for future laboratory experiments.

Critical assessment

Strengths and limitations

Computer predictions onlyClear limitation

The study measures no effect. It assumes that miRNAs from milk are taken up and regulate genes in human cells – an assumption that, according to the authors themselves, is scientifically disputed. No statements about effects in humans can be derived from it.

Only four donkey samplesClear limitation

The donkey data are based on four data sets from two earlier projects, one of them with a very low mapping rate (6.68 %). Breed, lactation stage and processing differ between the source data sets and are listed only in the appendix.

Conserved miRNAs onlyNote

Because all reads were evaluated against human references, donkey- or ruminant-specific miRNAs are missing. According to the authors, this means that the species-specific differences in particular are underestimated.

Mixed source samplesNote

The data sets include mature milk, colostrum (ten of the 13 cow samples) and milk fractions or milk cells, isolated with different methods. The data availability statement also mentions three GEO series that are not assigned in the sample description of the methods section.

Self-defined thresholdsNote

Categories are created via freely chosen limits (1.5-fold difference, 60th percentile). Mean scores and fold changes in Table 1 are based only on the 50 strongest genes per category, the gene counts on all 4577 genes.

Open and verifiableStrength

All source data are public, and intermediate results and code were provided. The authors test their classification with permutation and sensitivity tests and themselves name biases in the enrichment analysis.

clear limitationnotestrength

Editorial assessment: The work is a computational analysis of previously published data and contains no milk samples of its own. It describes which conserved miRNAs were frequent in the data sets analysed and how computed target gene profiles differ between types of milk. No statements about effects on genes, cells or human health can be derived from it. To the research archive.

Research question for DON DONKEY’S®

Can the miRNA abundances from public donkey milk data sets be confirmed with freshly collected samples from different breeds and lactation stages and with donkey-specific reference sequences?

microRNAComputational analysisMilk type comparison

Original publication

Conserved Core and Species-Specific Signatures in the Milk Exosomal microRNA Targetome: A Preliminary Comparative In Silico Analysis of Human, Cow, Goat and Donkey Milk

Original title of the publication.

Zoziuk M., Djibagaou A.D., Terrinoni A., Koroliouk D., Colizzi V.
International Journal of Molecular Sciences · 2026 · 27(15), 6952

Department of Experimental Medicine, University of Rome Tor Vergata, Rome, Italy; Laboratoire des Grandes Epidemies Tropicales, Faculty of Medicine, University Hospital Complex “Bon Samaritain”, N’Djamena, Chad

10.3390/ijms27156952: 10.3390/ijms27156952

Transparency: Italian Cooperation Agency (AID 12582); additional support from the Interdepartmental Centre for Comparative Medicine, Alternative Techniques and Aquaculture (CIMETA) of the University of Rome Tor Vergata and the Faculty of Medicine of the Good Samaritan Teaching Hospital in N’Djamena, Chad. The authors declare no conflicts of interest. No product-related statement for DON DONKEY’S®.

Summarised and reviewed by the DON DONKEY’S editorial team on the basis of the original publication · as of October 2026

End of the study summary

Note on our products

The study summarised above was carried out independently of DON DONKEY’S® and does not relate to our products. It serves solely to inform about the state of research.

DON DONKEY’S® donkey milk and donkey milk powder are foods. They are not intended to prevent, treat or cure any disease and do not replace medical advice.

From our shop

DON DONKEY’S® donkey milk

100 % donkey milk · raw milk, frozen

On this page we report the findings of a published study. Independently of that, you will find our donkey milk here: 100 % donkey milk, frozen as raw milk.

Transparency: The study examined milk from another farm – not the milk from our partner farms.

€24,90

100 ml bottle / 100ml · Unit price: €249,00 / 1 l · incl. VAT, plus shipping

View all sizes & purchase options →

Also in our shop

Donkey Milk Powder

Donkey milk in powder form

For anyone who prefers a longer shelf life: our donkey milk in powder form – easy to store and portion.

Transparency: The study did not examine our donkey milk powder.