Research Article | 06 Nov 2025

Machine learning-driven analysis of feed additives and intestinal microbiota diversity in broiler chickens: Clustering of mineral profiles and predictive diet modeling

Lyubov Sergeevna Grishina1 , Arthur Yurievich Zhigalov1 , Irina Pavlovna Bolodurina1 , Alexander Evgenievich Shukhman1 , Pavel Leonidovich Niryan1 , Olga Vilorievna Kvan2 , and Elena Vladimirovna Sheida2 Show more
VETERINARY WORLD | pg no. 3390-3408 | Vol. 18, Issue 11 | DOI: 10.14202/vetworld.2025.3390-3408
Citations: —
View PDF

Cite this Article

  • APA
  • MLA
  • Chicago
  • Vancouver
  • Harvard

              
            

          

        

      


      


      

      

        

          
Abstract

          
Background and Aim: The gut microbiota of broilers plays a pivotal role in nutrient absorption, immune modulation, and mineral metabolism. Feed additives can influence these microbial and physiological processes, yet their integrated effects remain insufficiently understood. This study aimed to intelligently evaluate the impact of various feed additives on the intestinal microbiota and mineral composition of broiler chickens and to develop machine learning (ML) models for clustering and classification of diet-associated mineral and microbial profiles.
Materials and Methods: A total of 385 Arbor Acres broilers (7 days old) were allocated into 11 groups, including one control semi-synthetic diet (SSD), one group with a semi-synthetic deficient diet (SSDD), and nine experimental groups receiving SSDD with different additives: Probiotics (Soya-bifidum and Sporobacterin), dietary fibers (cellulose, lactulose, and chitosan), enterosorbents (enterosgel and activated carbon), and ultrafine particles (UFPs) (Cu and Fe). Microbiota composition was assessed by 16S ribosomal RNA sequencing, and body mineral composition was determined through inductively coupled plasma mass spectrometer. To overcome data scarcity, synthetic records were generated using conditional tabular generative adversarial networks. K-means and hierarchical agglomerative clustering were used for mineral profile grouping, while logistic regression, SVM, and decision tree models classified diet types.
Results: Hierarchical clustering revealed six distinct mineral profile groups (Silhouette = 0.524), with SSD and SSDD forming separate clusters. Feed additives such as UFPs, chitosan, and activated carbon induced similar mineral patterns. Key differentiating biomarkers were cobalt, zinc, strontium, arsenic, and lithium (p < 0.05). The decision tree classifier achieved 74% accuracy in predicting diet types based on microbiota data. Alpha diversity analysis showed enhanced microbial richness in groups fed lactulose, enterosgel, cellulose, or activated carbon.
Conclusion: ML effectively elucidated complex relationships between diet, microbiota composition, and mineral metabolism in broilers. The integration of clustering and predictive models demonstrates the feasibility of intelligent feeding systems tailored to optimize gut health and nutrient utilization. Future studies integrating multi-omics data and broader farm-level validation will strengthen precision nutrition frameworks for sustainable poultry production.
          
Keywords: broilers, clustering, conditional tabular generative adversarial networks, decision tree, feed additives, gut microbiota, machine learning, mineral metabolism.