Background: Fetal growth restriction (FGR), the largest contributor to fetal death, is associated with neonatal morbidity, suboptimal neurodevelopment, and chronic diseases later in life. Large variability in clinical presentation and perinatal risks have made it difficult to implement a clinical classification for FGR. Machine learning (ML) has the potential to revolutionize clinical decision-making in the management of FGR by identifying new phenotypes based on multi-omics data.Methods: An unsupervised ML method - similarity network fusion (SNF) - was used to analyze 546 pregnancies complicated with FGR by integrating clinical characteristics, angiogenic factors, ultrasound, and 1H-NMR based metabolomics. Data integration, cluster determination, and biological groups were obtained. Finally, ROC curves were used to compare the diagnostic accuracy of SNF with the current clinical classification.Findings: Two FGR molecular subtypes were identified: a high-risk (cluster A: 37·6%) and a low-risk group (cluster B: 62·4%). Compared to cluster B, fetuses in cluster A debuted earlier in gestation, had a shorter gestational length and a specific metabolic signature, and presented higher rates of preeclampsia (55·1% vs. 8·9%, P<0·001), perinatal deaths (6·94%, vs. 0%, P<0·001), and suboptimal neurodevelopment. Clusters generated by SNF significantly outperformed single data subtype analysis and the current clinical classification in the prediction of adverse maternal and neonatal outcomes.Interpretation: Our approach can be leveraged to develop classification systems for FGR based on molecular and clinical signatures rather than expert consensus. This aids new insights into the pathobiology and solid grounds for improving the prediction of adverse outcomes in suboptimal fetal growth.Funding Information: This project has been funded with the Erasmus + Programme of the European Union (Framework Agreement number: 2013-0040). This publication reflects the author's views only, and the Commission cannot be held responsible for any use of the information contained therein. Additionally, the research leading to these results has received funding from the “Instituto de Salud Carlos III (PI17/00675, PI18/00073, PI20/00246, PI20/00741,INT21/00027)” integrated by “Plan Nacional de I+D+I” and financed by “ISCIII-Subdirección General de Evaluación” and “Fondo Europeo de Desarrollo Regional (FEDER) “Una Manera de hacer Europa,” Cerebra Foundation for the Brain Injured Child (Carmarthen, Wales, UK), ASISA foundation, Department de Salut (ICT), and CERCA Programme from Generalitat de Catalunya. In addition, JM was supported by a predoctoral governmental “Bolivar Gana con Ciencia” grant from Colombia, while CP was supported by a Juan Rodes Grant from “Instituto de Salud Carlos III” (JR19/0006).Declaration of Interests: The authors declare no competing interests.Ethics Approval Statement: The ethics committee of (HCB)-Institutd’Investigacions Biomédiques August Pi i Sunyer (IDIBAPS), Hospital Clinic de Barcelona, (review board 2014/7154). Written informed consent was obtained from each participant. INCOMPLETE.