In general, similar considerations and precautions should be applied for food samples subjected to NGS and to conventional microbiological analyses to ensure sample representativeness and integrity. The quantity of samples and repetitions can significantly impact data accuracy and reproducibility. Hence, researchers need to balance the desire for large repetitions and the costs and feasibility of processing these samples. This is especially challenging for food sampling, as raw material microbiota can vary greatly among samples and can change during processing (Sadurski et al., 2024). Both probability (random) and non-probability (non-random)-based approaches have been proven effective in designing sampling schemes tailored to the research question and food type (Taherdoost, 2016).

For food samples to be analyzed using NGS, changes in cell cultivability is not a major concern, and therefore, harsher conditions can be applied, such as snap freezing, rapid drying, or even certain chemical preservatives can be applied to prevent continued microbial growth or other changes that alter the sample nucleic acid profiles (Fricker et al., 2019).

Proper storage is important to prevent nucleic acid degradation or microbial growth (Kazantseva et al., 2021). To mitigate these risks, food samples are usually cooled to 4 °C or frozen at −20 °C or −80 °C, depending on the available facilities (Lear et al., 2018). In contrast, shelf-stable products like freeze-dried cheonggukjang, a fermented soybean food of Korea, can be stored at room temperature in moisture-free conditions, due to their inherent microbial and chemical stability (Tamang et al., 2022).

Sampling must also account for different processing stages, which influence microbial composition. For instance, Barcenilla et al. (2024) used pooled swabs to improve microbial recovery from raw meat surfaces, while tailoring sampling techniques such as direct sampling or surface swabbing for various end products like fermented sausages or cured meats. Similarly, studies like Gaire et al. (2024) and Sequino et al. (2024) have emphasized the importance of stage-specific sampling to capture shifts in microbial diversity and antimicrobial resistance.

Ultimately, successful NGS-based food science studies rely on thoughtful sampling design and proper storage. Researchers should tailor these steps to the specific food matrix, processing stage, and study goals to ensure meaningful and reproducible results.

Nucleic acid extraction from food matrices is a vital process in NGS of food products enabling the detection and study of genetic material (Tan and Yiap, 2009). Nucleic acid extraction methods consist of three steps: lysis, purification, and nucleic acid recovery. Furthermore, extraction can be performed using conventional protocols or commercially available kits, with the choice of method depending on sample complexity and study objectives (Sajali et al., 2018).

Cell lysis is carried out to break open microbial cells and release nucleic acids. Lysis methods can be chemical, enzymatic, mechanical, or a combination, depending on the complexity of the matrix (Lever et al., 2015). For instance, enzymatic lysis combined with mechanical disruption has been effectively applied to romaine lettuce (Gu et al., 2024) and coffee (Pothakos et al., 2020) illustrating the importance of method customization based on the nature of the sample.

After cell lysis, nucleic acid purification is performed to separate DNA or RNA from cellular debris, proteins, and inhibitors, typically using liquid-liquid (LLE) or solid-phase extraction methods (SPE) (Ruggieri et al., 2016). For example, the CTAB and chloroform LLE method was used by Lima et al. (2022) to isolate DNA from fermented cacao and by (Xiao et al., 2022) for RNA isolation from Sichuan paocai, highlighting the continued relevance of traditional reagents in specific contexts. SPE kits utilizing silica-based filters reduce reliance on organic solvents and enhance efficiency in nucleic acid recovery (Emaus et al., 2020). These kits have been applied to extract DNA from raw meat swabs (Barcenilla et al., 2024; Gaire et al., 2024) and fermented soybean products (Kharnaior and Tamang, 2022; Tamang et al., 2022) showing their versatility. However, effectiveness can vary depending on sample composition, emphasizing the need for further optimization, particularly for high-fat or polyphenol-rich matrices.

Following lysis and purification, nucleic acid recovery is carried out by pelleting the nucleic acids through centrifugation, followed by resuspension in a suitable buffer. Nuclease-free water or 1× TE buffer is commonly used, with TE offering added protection (Schenk et al., 2023).

In conclusion, nucleic acid extraction is crucial for NGS-based studies on food science, employing diverse methods to ensure DNA/RNA quality and quantity. Tailoring techniques to specific samples enhances analytical reliability and accuracy, reinforcing confidence in research findings. Table 2 provides an overview of nucleic acid extraction techniques utilized in food science.

NGS in food science often incorporates a PCR amplification step to selectively enrich target marker genes from complex mixtures of genomic DNA. For instance, 16S ribosomal RNA (rRNA) gene sequencing is a targeted approach where PCR is used to amplify specific hypervariable regions of the 16S rRNA gene from diverse bacterial populations present in food samples (Abellan-Schneyder et al., 2021). This amplification enables in-depth profiling of microbial communities, which is essential for assessing food safety, quality, and shelf-life.

In addition to the 16S rRNA gene, other marker genes such as the internal transcribed spacer (ITS) region for fungal identification and the 18S rRNA gene for eukaryotic microorganisms are also routinely targeted for amplicon sequencing (Banos et al., 2018). By selectively amplifying these marker genes, researchers can generate high-quality, focused libraries for NGS, ultimately providing deeper insights into the microbial ecology of food products and aiding in effective monitoring of potential contaminants (Fagerlund et al., 2021).

Additionally, marker genes play a crucial role in food authenticity determination by enabling the detection of species-specific DNA sequences to verify the origin and composition of food products (Vishnuraj et al., 2023). Commonly used markers include mitochondrial genes such as cytochrome b (cyt b) and cytochrome c oxidase I (COI) for identifying animal species, as well as nuclear genes like the rbcL and matK genes for plant species authentication (Grazina et al., 2020; Afifa et al., 2021; Zhao Z. et al., 2024). The cyt b and COI genes encode proteins involved in mitochondrial respiration and are favored for their high interspecies variability (Antil et al., 2023; Tutuş et al., 2025). In plants, rbcL encodes the large subunit ribulose-1,5-bisphosphate carboxylase/oxygenase (RuBisCO) essential for photosynthesis, while matK encodes maturase K, involved in chloroplast RNA splicing and known for its high substitution rate and discriminatory power at the species level (Jia et al., 2025). These genetic markers help prevent food fraud, ensure label accuracy, and protect consumers from misrepresentation or adulteration of food products.

In addition to targeted gene amplification, some NGS applications in food science, such as shotgun metagenomics, incorporate optional PCR amplification during library preparation to enhance sequencing efficiency. For example, the Nextera XT DNA Library Preparation Kit employs enzymatic fragmentation followed by PCR amplification to generate sequencing-ready libraries, particularly when working with food samples containing low-abundance of target DNA (Kim et al., 2013). This approach facilitates comprehensive microbial profiling and functional gene analysis in food samples. However, as seen in metagenomic studies like the genetic characterization of dengue viruses (Lizarazo et al., 2019), excessive PCR cycles could introduce additional biases and duplicates, potentially affecting sequencing accuracy.

NGS library preparation involves processing nucleic acids (RNA or DNA) to align with the chosen sequencing platform (Head et al., 2014). This includes fragmentation, multiplexing, normalization and adapter ligation (Abdi et al., 2024). While amplicon sequencing relies on short DNA fragments of target regions selectively amplified via PCR, shotgun sequencing requires fragmentation of total DNA using sonication, enzymatic digestion, or mechanical shearing (Srinivas et al., 2022). On the other hand, for RNA sequencing in metatranscriptomic analyses, mRNA molecules are converted to cDNA fragment by reverse transcription, using either poly(A) tailing (targeting eukaryotes) or random priming (targeting prokaryotes) (Saliba et al., 2014). Short-read platforms like Illumina favor fragments under 450 bp, whereas long-read platforms such as PacBio and Oxford Nanopore Technologies (ONT) require high molecular weight (HMW) DNA, sometimes sheared to ∼20 kb for improved sequencing efficiency (Madoui et al., 2015; Ribarska et al., 2022).

Multiplexing, or indexing, allows multiple libraries to be pooled and sequenced together using unique index sequences. To ensure equal representation, libraries must first be normalized to the same DNA/RNA concentration before being pooled in equal volumes (Muller et al., 2019). However, challenges like index hopping (where reads are misassigned) can introduce errors, particularly in Illumina platforms with patterned flow cells. Unique dual indexing and nested metabarcoding help mitigate these issues (MacConaill et al., 2018; Guenay-Greunke et al., 2021).

Adapter ligation attaches platform-specific sequences to DNA fragments, enabling binding to flow cells. Illumina uses adapters for anchoring, whereas PacBio employs hairpin adapters for circular DNA sequencing. ONT ligates adapters that guide DNA strands through nanopores (Oikonomopoulos et al., 2020; MacKenzie and Argyropoulos, 2023). Amplicon sequencing on Illumina integrates adapters directly during PCR, bypassing the ligation step (Glenn et al., 2019).

Minimizing contamination is critical, as microbial DNA from reagents can bias NGS analyses. Therefore, proper laboratory practices, negative controls, and mock microbial communities help reduce this risk (Salter et al., 2014). Advances in automation and quality control continue to enhance the reliability and efficiency of NGS library preparation in food science applications.

NGS methods for food microbiome analysis can be broadly divided into targeted (amplicon-based) and random (shotgun) sequencing approaches, each serving different analytical goals (Sekse et al., 2017).

Amplicon sequencing, also known as targeted sequencing, focuses on specific genetic markers, such as the 16S rRNA gene for bacteria, ITS regions for fungi, and COI genes for metazoans (Francioli et al., 2021). This approach, using platforms like Illumina MiSeq and Ion Torrent PGM is widely adopted due to cost-effectiveness and established analytical pipelines (Lira et al., 2020; Syromyatnikov et al., 2020; Anagnostopoulos et al., 2022; 2023; Güley et al., 2023). However, PCR-induced biases in primer selection can significantly affect microbiota characterization (Sergeant et al., 2012). In fungi, ITS length variation can distort community structure by favoring shorter fragments (De Filippis et al., 2017). Amplicon sequencing also lacks strain-level resolution, limiting food safety assessments (La Reau et al., 2023). To mitigate biases, researchers can consider alternative fungal-specific targets, such as 26S and 18S rRNA genes, for improved accuracy (Banos et al., 2018; Mota-Gutierrez et al., 2019).

Shotgun metagenomic sequencing, in contrast, sequences all DNA present in a sample, providing a comprehensive view of the genetic content, including functional genes, metabolic pathways, and antimicrobial resistance determinants (Tyagi et al., 2019). This approach enables deep characterization of food microbiomes, making it ideal for studying fermentation and resistome profiling (Pothakos et al., 2020; Kim et al., 2021; Kharnaior and Tamang, 2022; Lima et al., 2022; Tamang et al., 2022; Lee et al., 2024). However, it cannot easily distinguish live from dead cells. When viability is critical, combining metagenomics with culture-dependent techniques ensures accurate risk assessment (Lindner et al., 2024). Shotgun metagenomics requires robust computational resources for assembly and annotation (Tremblay et al., 2022). Platforms like Illumina NovaSeq and PacBio Sequel II offer high-throughput sequencing, while long-read technologies such as ONT and PacBio HiFi enhance genome assembly and resolution in complex food matrices (Zhang et al., 2022).

In addition, whole genome sequencing (WGS) and RNA sequencing (RNA-Seq) are also important for microbial analysis and food safety. Platforms used for WGS of foodborne pathogens include Oxford Nanopore MinION, for Escherichia coli characterization from meat products (Wang et al., 2024; Martínez-Álvarez et al., 2025). Illumina systems like NextSeq, HiSeq and NovaSeq, are widely employed for RNA-Seq for elucidating important microbial functions during food fermentation processes, as well as in transcriptomic studies of foodborne pathogens in fresh produce and related matrices (Xiao et al., 2021; Redding et al., 2024; Xiong et al., 2024; 2024; Ding et al., 2025). Comprehensive metatranscriptomic analyses of mRNA dynamics can also provide insights on microbial interactions in complex microbial communities or biological processes.

To enhance sequencing efficiency and accuracy, several strategies can be employed. Optimizing sample preparation helps minimize biases, especially in complex food matrices. Choosing the right sequencing platform ensures a balance between cost, read length, and throughput. Combining amplicon and shotgun approaches can provide complementary insights, and advancements in bioinformatics tools as discussed in the following section, such as machine learning-based classification and improved genome binning, can facilitate improved data interpretation for better tracking of foodborne pathogens and microbial ecology shifts.

The vast amount of data generated by NGS requires robust bioinformatics pipelines for accurate analysis and interpretation. The bioinformatics workflow varies depending on the sequencing approach as well as the specific NGS platform used.

The application of NGS in food safety assessment has significantly improved pathogen detection, microbial community analysis, and AMR surveillance (Pennone et al., 2022).

Metagenetics, primarily based on 16S rRNA gene sequencing, is used to study microbial communities in food and their processing environments. For example, Anagnostopoulos et al. (2022) and Anagnostopoulos et al. (2023) used 16S metabarcoding to analyze microbial succession in ice-stored seabream, revealing shifts in bacteria like Pseudomonas and Shewanella associated with spoilage. Similarly, Syromyatnikov et al. (2020) applied 16S sequencing to butter microbiota, detecting opportunistic pathogens missed by traditional methods. Díaz Cárdenas et al. (2024) studied street-vended foods in Ecuador, identifying dominant genera like Acinetobacter, Lactococcus, and Vibrio. They found twenty-nine spoilage bacteria and twenty-four opportunistic pathogens, underscoring the food safety risks in these environments and highlighting the importance of metagenetics for food safety monitoring.

WGS offers unmatched precision for examining individual microbial isolates, allowing for meticulous strain tracking during outbreaks. Bai et al. (2024) applied WGS to fresh-cut fruits and vegetables, detecting Salmonella, Listeria monocytogenes, and E. coli along with virulence and resistance genes. Similarly, Habib et al. (2024) found multidrug-resistant Salmonella enterica in chilled broiler chicken. These bacterial pathogens are among the most relevant to public health in the various food production sectors in the EU/EEA (Koutsoumanis et al., 2024). Moreover, the emergence of long-read sequencing technologies like Oxford Nanopore suggests a pathway toward real-time, portable, on-site food safety testing (Lee et al., 2024). To provide a clearer overview of key foodborne pathogens, their typical food sources, and their detection using NGS, Table 4 summarizes selected examples aligned with recent research.

Beyond outbreak detection, NGS also strengthens food safety through its role in preventive quality control and source attribution. By characterizing microbial contaminants at high resolution, NGS can help detect the origin of contamination, differentiating whether it arose from raw ingredients, surfaces, or environmental niches. This technology is especially useful when multiple sources of contamination are involved, as genomic comparison of isolates from diverse sampling points can uncover distinct strain lineages or reveal shared transmission routes (Rantsiou et al., 2018; Koutsoumanis et al., 2024).

Metatranscriptomics, while less frequently deployed, introduces a unique analytical layer by targeting gene expression (RNA) rather than solely genetic potential (DNA). This approach is particularly useful for assessing spoilage mechanisms and microbial metabolism in perishable foods. For example, de Lira et al. (2020) integrated metatranscriptomics to examine histamine production in fish, which highlights its utility in connecting specific microbial functions to observable food quality deterioration. Its application in verifying the activity of foodborne viruses like norovirus Yang et al. (2024), also demonstrates its value extends beyond bacteria. This technique, therefore, provides critical insights into which microbes are metabolically active and the functions they are executing within the food matrix at a specific moment.

The widespread adoption of NGS across varied food industries from dairy (Ma et al., 2018; Yang et al., 2020; Liang et al., 2022) and meat/poultry products (Barcenilla et al., 2024; Martínez-Álvarez et al., 2025; Wang et al., 2024) to fresh produce (Solcova et al., 2021; Gu et al., 2024) attests to its flexibility, but also brings common challenges into focus. A consistent theme emerging from these diverse studies is the significant influence of processing environments on microbial ecology and the spread of AMR genes (Gaire et al., 2024; Koutsoumanis et al., 2024), alongside the risks posed by pathogens and resistance elements in raw ingredients. This implies that microbial risks are ubiquitous yet shaped by the specifics of each food type and its journey through the production chain.

For effective management, NGS monitoring strategies should be customized for different sectors, concentrating on critical control points identified via thorough environmental and product analyses. Promoting the sharing of data and insights from these varied applications, following the example of successful collaborative projects (Allard et al., 2016) will be instrumental in refining best practices and informing risk management universally.

Comparing NGS with conventional methods starkly illustrates its advantages in speed, resolution, and culture-independent analysis, facilitating more potent outbreak responses and surveillance (Forbes et al., 2017; Hilt and Ferrieri, 2022; Panwar et al., 2023). The practical impact is evident in real-world scenarios, such as the FDA resolving complex outbreaks using WGS data in regulatory decisions (Ottesen and Ramachandran, 2019; Mak et al., 2025). Nevertheless, significant hurdles impede its routine implementation: costs, the demand for specialized bioinformatics skills and absence of standardization present considerable barriers. Future progress hinges on collaborative initiatives aimed at standardizing protocols and bioinformatics workflows, substantial investment in training and infrastructure, and the creation of unambiguous guidelines for validation and data interpretation.

Food fermentation enhances food products through complex microbial activities, traditionally studied with methods that were limited in scope. NGS offers a transformative approach, providing extensive insights into microbial community structure, function, and dynamics (Pothakos et al., 2020; Ferrara et al., 2021; Kim et al., 2021; Lima et al., 2022; Tamang et al., 2022; Xiong et al., 2024). The application of NGS reveals the previously hidden microbial diversity, including unculturable organisms, vital for fully understanding fermentation processes. Figure 2 summarizes the main contributions of NGS to food fermentation, including insights into microbial dynamics, starter culture development, process optimization, and the linkage between gene expression and flavor compound formation.

NGS methods like amplicon sequencing and shotgun metagenomics provide detailed taxonomic profiles (Ferrara et al., 2021; Tigrero-Vaca et al., 2022; Vaccalluzzo et al., 2022), and metatranscriptomics identifies active genes and microbes during fermentation stages (Pothakos et al., 2020; Kim et al., 2021; Xiong et al., 2024). Studies exemplify this power: identifying biogenic amine producers in soy sauce highlights NGS’s role in assessing safety aspects alongside flavor development (Kim et al., 2021); revealing microbial shifts influencing volatile compounds in olives demonstrates its utility in targeted quality improvement (Vaccalluzzo et al., 2022); and uncovering complex networks in focaccia contrasts sharply with the limited view from plating methods (Ferrara et al., 2021).

The consistent finding across fermentation studies is that NGS offers a far more comprehensive view than traditional methods, which is crucial for informed development of starter cultures. A potential gap suggested by the focus on dominant bacteria (like lactic acid bacteria) is the need for deeper investigation into the roles of sub-dominant bacteria, yeasts, and potentially other microbes; employing deeper sequencing or targeted enrichment could address this.

Functionally, metagenomics predicts the metabolic potential (Kharnaior and Tamang, 2022; Lima et al., 2022; Tamang et al., 2022), while metatranscriptomics confirms active pathways (Kim et al., 2021; Xiong et al., 2024). Linking specific microbial gene expression to metabolite formation provides direct targets for intervention (Kim et al., 2021; Kharnaior and Tamang, 2022; Xiao et al., 2022; Xiong et al., 2024). Uncovering enzymatic functions in cacao or amino acid profiles in cheonggukjang directly informs strategies for enhancing specific sensory attributes (Lima et al., 2022; Tamang et al., 2022; Tigrero-Vaca et al., 2022). A significant implication is the ability to move from spontaneous to controlled fermentations by understanding and manipulating these functional links.

Ultimately, the application of NGS profoundly shapes the future of food fermentation. Identifying key microbes and metabolic pathways facilitates the rational design and selection of starter cultures tailored for specific outcomes, such as desired flavor profiles in cacao or controlled activity in coffee (Pregolini et al., 2021; Lima et al., 2022; Tigrero-Vaca et al., 2022; Constante Catuto et al., 2024). Understanding microbial succession and function, as shown in paocai and suancai, is critical for process optimization (Xiao et al., 2022; Xiong et al., 2024). A clear direction arising from these studies is the synergy gained from multi-omics approaches; combining genomics, transcriptomics, and metabolomics offers the most holistic view (Kharnaior and Tamang, 2022; Xiong et al., 2024) and represents a key strategy to address remaining knowledge gaps about complex microbial interactions and their precise impact on final product characteristics.

In addition to food safety and fermentation, NGS is driving innovative approaches in food authentication, traceability, and product integrity. Techniques like metagenomics and metabarcoding are being employed to ensure the quality and authenticity of food products.

Metabarcoding, is a molecular technique that enables the identification of multiple species within a mixed sample, such as bulk or environmental DNA, through high-throughput sequencing of a targeted genetic marker (Liu et al., 2020). In food science, metabarcoding plays a crucial role in species authentication, detection of adulteration, and monitoring of microbial communities (Mottola et al., 2024). For example, Detcharoen et al. (2024) used Nanopore sequencing to authenticate fish species in surimi-based products, offering a rapid and accurate method for detecting species substitution and ensuring food traceability. Additionally, Giusti et al. (2024) employed metabarcoding to uncover species mislabeling in insect-based products marketed in the EU, underscoring its value in ensuring transparency and regulatory compliance in novel food markets.

Nanopore sequencing has also proven valuable in real-time, on-site detection of contaminants in brewing processes. Kurniawan et al. (2021) developed a rapid nanopore-based platform to identify beer-spoiling bacteria directly in breweries, improving quality control and minimizing spoilage risks. In another brewing application, (Shinohara et al., 2021), utilized nanopore sequencing to identify yeast species quickly and accurately in breweries, enhancing yeast management and beer production.

Beyond these applications, NGS has also emerged as a powerful tool in studying food microbiome interactions. For example, Lerma-Aguilera et al. (2024) examined how cooking methods influence microbial diversity, while Küçükgöz et al. (2025) examined the impact of fermented beetroot ketchup on gut microbiota. Similarly, Gong et al. (2024) investigated the effects of processed oats and pinto beans on gut microbiota, demonstrating that dietary fibers from these foods promote beneficial bacteria. Additionally, Milani et al. (2025) demonstrated how bacteria present in cheese can modulate the gut microbiome, emphasizing the role of food as a vehicle for beneficial microbes. These findings underscore the growing importance of NGS in assessing the functional effects of food on human health.

The convergence of NGS with AI and ML is rapidly reshaping food science, offering predictive power and real-time decision-making in areas such as food safety, traceability, and microbial risk assessment (Liao et al., 2024). These approaches go beyond traditional detection and profiling, enabling the extraction of meaningful patterns from complex, high-dimensional sequencing data.

Supervised learning models, such as random forests and decision trees, have been used to predict foodborne pathogen contamination. For example, Bolinger et al. (2021) applied random forest classifiers to 16S rRNA microbiome profiles from poultry rinsates to estimate Salmonella contamination risk at slaughter. Their model identified specific microbial signatures as reliable proxies for pathogen presence, demonstrating the predictive value of sequencing data integrated with ML algorithms.

Explainable AI approaches are also emerging. Ince et al. (2025) combined WGS with a decision-tree model, using SHAP (SHapley Additive exPlanations) values to quantify the influence of microbiota features and temporal variables on the growth of Clostridium perfringens in pork. This interpretable framework offered microbiologically relevant insights, advancing food spoilage prediction and process control.

Baker et al. (2023) further demonstrated the use of ML in food safety surveillance by combining shotgun metagenomics with unsupervised clustering and supervised models to examine microbial communities and AMR genes across poultry farms and abattoirs in China. The study revealed consistent resistome profiles across sites, suggesting common selective pressures and potential routes of cross-contamination. This illustrates how AI-enhanced NGS analyses can support comprehensive surveillance of AMR and microbial hazards across the food production chain.

Beyond safety, ML tools have also been applied to food authentication. Sabater et al. (2024) combined metagenomic sequencing with supervised machine learning to trace the geographic origin of Spanish Protected Designation of Origin (PDO) honey, identifying microbial community signatures as indicators of authenticity. This approach underscores the potential of ML-integrated NGS data to support fraud detection and ensure product traceability. Similar strategies could be extended to other high-value or regulated food products, making AI-enhanced NGS an asset in modern food authentication frameworks.

In the domain of food fermentation, ML and multi-omics approaches are proving especially powerful (Li et al., 2025). integrated metagenomics sequencing and metabolomics with machine learning algorithm (logistic regression, K-nearest neighbors (KNN), and random forest) to classify abnormal stacking fermentations in sauce-flavor Baijiu. Their models successfully distinguished between types of fermentation failure, and SHAP analysis was used to identify key microbial and metabolite biomarkers driving the predictions. This study exemplifies how AI enhanced multi-omics can uncover mechanistic insights into microbial dynamics and functional disruptions, supporting both quality control and the optimization of traditional fermentation processes.

Figure 3 presents a schematic representation of AI and ML applications in food science enabled by NGS, including pathogen prediction, AMR tracking, fermentation monitoring, and food authentication. The diagram highlights the integration of models such as random forest, SHAP, and supervised learning with omics data to support predictive and diagnostic capabilities across food systems.

Despite these advances, challenges remain. Many models are dataset-dependent, limiting generalizability across food systems. The need for standardized pipelines, validated microbial markers, and robust training datasets is critical. Furthermore, regulatory acceptance and interpretability of AI outputs are ongoing concerns.

Looking forward, integrating multi-omics data (e.g., genomics, transcriptomics, metabolomics) with hybrid AI models may provide a more holistic understanding of microbial behavior, food spoilage mechanisms, and safety risks. As sequencing technologies and computational methods evolve, AI and ML will become increasingly central to precision food safety, real-time diagnostics, and intelligent quality assurance.