This study aims to enhance treatment protocols for pediatric ulcerative colitis (UC), an inflammatory bowel disease, by identifying microbial biomarkers associated with relapse in treatment-naive. Current first-line treatments, such as 5-ASA drugs and corticosteroids, are effective in only half of the patients. Identifying those at risk of relapse could enable early introduction of more aggressive treatments, like immunosuppressants. Biopsies from 48 paediatric patients (2-16 years of age) were collected during colonoscopy. Of these, 23 experienced relapse within 6 months (PUCAI 10+). Next-generation sequencing of V3-V4 hypervariable region and 16S qPCR provided data on relative and absolute bacterial abundances. Taxa differential abundance between relapse and remission groups was identified with zero-inflated Gaussian and negative binomial mixed-effect models. Meta-variables relevance to microbiome composition was assessed with VpThemAll package. XGBoost-based machine-learning models were used to predict relapse. Diversity and species richness were diminished in the relapse group. 23 and 10 differential taxa were differential in negative binomial and Gaussian models respectively. Differential taxa included probiotic bacteria (Bifidobacterium genus, elevated in remission group) and pro-inflammatory species (from Veilonella and Fusobacterium genus, elevated in relapse). Patient treatment status ranked as third most relevant variable in explaining variance in the microbiota. Models based on ascending colon samples microbiome and demographic data reach AUC above 0.7. Gut microbial composition is linked to treatment response in paediatric UC. Differential taxa include probiotic bacteria (Bifidobacterium genus, elevated in remission group) and pro-inflammatory species (from Veilonella and Fusobacterium genus, elevated in relapse). Microbiome-based models achieve good performance in predicting relapse.