scVIP: personalized modeling of single-cell transcriptomes for developmental and disease phenotypes

Single-cell RNA sequencing reveals cellular heterogeneity, but linking cellular states to individual-level phenotypes remains challenging. We present scVIP, a generative framework that integrates transcriptional profiles and phenotypic markers to learn personalized individual-level embeddings using generative models and cell-type-aware multi-instance learning. scVIP predicts developmental age, disease progression, and neuropathology, while harmonizing datasets with distinct phenotype definitions. The model highlights disease-relevant cell populations and transcriptional programs underlying neurodegeneration.