Stem cell fate prediction improved with purity-aware deep learning
PuriGen: Purity-Aware Deep Generative Modeling for Predicting Stem Cell Lineage Fate
Computational Engineering, Finance, and Science
Summary
Predicting how stem cells develop into different types of cells is important for tissue healing and regeneration. Existing methods struggle with handling mixed cell types and how cells change over time. The authors develop PuriGen, a deep learning system that uses special techniques to better understand cell mixtures and timing from gene data. Their approach works better than previous methods on data from stem cells grown on various materials and at different stages. This can help improve how scientists study and guide stem cell development.
What this means in practice
- •For biotech engineers: Predict stem cell development outcomes under different biomaterials to optimize tissue engineering protocols.
- •For pharmaceutical data teams: Improve analysis of bulk gene expression data to better assess drug effects on stem cell differentiation.
Authors
Yanli Li, Tianying Sheng, Hala Zreiqat, ZuFu Lu, Zhiyong Wang
Abstract
Predicting stem-cell lineage fate is central to elucidating biomaterial-stem cell interactions and optimizing strategies for tissue regeneration. While existing approaches demonstrate the feasibility of lineage prediction, they remain limited in modelling compositional heterogeneity and time-dependent differentiation dynamics. To mitigate this gap, we propose PuriGen, a purity-aware deep generative framework for biomaterial-induced lineage prediction from bulk transcriptomic data. PuriGen consists of two key modules, PuritySCVI and PuritySCANVI, which extend scVI and scANVI by incorporating a pretrained GBMPurity head to provide a sample-level purity-like compositional signal. In addition, purity-aware regularisation and temporal constraints are introduced to encourage informative purity prediction and biologically consistent latent representations. Experiments on bulk mesenchymal stem-cell datasets collected across multiple biomaterial conditions and induction stages show that the proposed framework achieves better performance than the existing methods.