Stem cell therapy emerges as an effective approach for treating various currently untreatable diseases. However, fatal and unknown risks caused by their systemic use remain to be a major obstacle to clinical application. We developed a functional single-cell RNA sequencing (scRNA-seq) procedure and identified that transcriptomic heterogeneity of adipose-derived stromal cells (ADSCs) in cultures is responsible for a fatal embolic risk of these cells in the host. The pro-embolic subpopulation of ADSCs in cultures was sorted by gene set enrichment analysis (GSEA) and verified by a supervised machine learning analysis. A mathematical model was developed and validated for the prediction of embolic risk of cultured ADSCs in animal models and further confirmed by its application to public data. Importantly, modification of culture conditions prevented the embolic risk. This novel procedure can be applied to other aspects of risk assessment and would help further the development of stem cell clinical applications. Keywords: Stem cells research; computational bioinformatics; transcriptomics.