The remarkable capacity of the liver to regenerate its lost mass after resection makes living donor liver transplantation a successful treatment option. However, donor heterogeneity significantly influences recovery trajectories, highlighting the need for individualized monitoring. With the rising incidence of liver diseases, safer transplant procedures and improved donor care are urgently needed. Current clinical markers provide only limited snapshots of recovery, making it challenging to predict long-term outcomes. Following partial hepatectomy, precise liver mass recovery requires tightly regulated hepatocyte proliferation. We identified distinct gene expression patterns associated with liver regeneration by analyzing blood-derived gene expression measurements from twelve donors followed over a year. Using a deep learning-based framework, we integrated these patterns with a mathematical model of hepatocyte transitions to develop a personalized, progressive mechanistic digital twin-a virtual liver model that predicts donor-specific recovery trajectories. Central to our approach is a mechanistically identifiable latent space, defined by variables derived from a physiologically grounded differential equation model of liver regeneration, which enables biologically interpretable, bidirectional mapping between gene expression data and model dynamics. This approach integrates clinical genomics and computational modeling to enhance post-surgical care, ensuring safer transplants and improved donor recovery.