472 - AI-Driven Characterization of Gut Microbial Maturation Trajectories in Preterm and Term Infants

Publication Number: 4463.472

Hyun-Kyung Park, Hanyang University College of Medicine, Seoul, Korea, Seoul, Seoul-t'ukpyolsi, Republic of Korea; Jae Kyoon Hwang, Hanyang university, seongdong gu, Seoul-t'ukpyolsi, Republic of Korea; Tae Hyun Kim, Hanyang University, Seoul, Seoul-t'ukpyolsi, Republic of Korea; Byong-Hun Jeon, Hanyang University, Gwangjin-gu District, Seoul-t'ukpyolsi, Republic of Korea

Background: The neonatal gut microbiome plays a crucial role in immune and metabolic maturation; however, its development differs substantially between preterm and term infants. Understanding these differences is challenging due to the high dimensionality and sparsity of microbiome data, as well as limited sample sizes.

Objective: To overcome these constraints, we applied an artificial intelligence (AI)-based machine learning framework to explore temporal microbial dynamics and community structures across neonatal groups.

Design/Methods: Fecal microbiome data from preterm, healthy term, and disease-affected term infants were processed through a structured analytical pipeline. After log transformation and taxonomic hierarchy alignment (Domain-Species), phylogenetically informed matrices were constructed for each sample. These matrices were transformed into image representations and analyzed using a ResNet-based convolutional neural network, followed by t-distributed stochastic neighbor embedding (t-SNE) for dimensional visualization. Serial sampling data were incorporated to track within-infant microbial trajectories over time.

Results: Distinct clustering patterns were observed among preterm, normal term, and disease-affected term infants. Preterm infants exhibited a unique and tightly clustered microbial configuration, indicating delayed or altered microbial maturation. Term infants showed broader and overlapping distributions, suggesting a more homogeneous progression of microbial colonization. Longitudinal tracking demonstrated progressive differentiation and maturation of microbial communities beyond 60 days of life.

Conclusion(s): This AI-based integrative approach successfully characterized microbial community dynamics and revealed distinct maturation trajectories between preterm and term infants. These findings highlight the potential of machine learning for identifying developmental biomarkers of the neonatal microbiome and for guiding early intervention strategies to promote healthy microbial and immune development.

t-SNE plot of ResNet-derived latent features of temporal microbiome trajectories among preterm, sick term, and term infants.
Preterm infants (red circle, group 2) formed a dense and clearly separable cluster, indicating a unique gut microbial configuration. Normal term (blue circle, group 1) and abnormal term (grey circle, group 0) infants showed broad and overlapping distributions, suggesting similar microbial profiles. The black arrow represents the direction of sampling time flow.

Trajectories of gut microbiome in individual preterm infants
Numbers denote patient code (first) and sampling day (second).