Machine learning and bioinformatics experts from different European countries and representatives from the PERMIT consortium met for an online workshop on 18 June 2021, to discuss the current challenges and limitations in machine learning (ML) analyses of omics data for patient stratification. Based on a scoping review of the relevant biomedical literature presented in the first part of the meeting, the participants collected experiences and opinions on common shortcomings in ML patient stratification projects, covering both methodological aspects (e.g. choice of domain-specific vs. generic approaches, focus on clinical variables vs. focus on molecular data) and practical aspects related to interdisciplinary collaborations (e.g. challenges in the communication between computational, experimental and clinical researchers). Possible mitigation strategies were discussed, and a structural categorization of identified common challenges and associated recommendations was proposed. Finally, a follow-up meeting was planned to prepare a final structured summary of the developed recommendations.