Donor-Specific Digital Twin for Living Donor Liver Transplant Recovery

Abstract

Liver resection initiates a meticulously coordinated hyperplasia process characterized by regulated cell proliferation that drives liver regeneration. This process concludes with the complete restoration of liver mass, showcasing the precision and robustness of this homeostasis. The remarkable capacity of the liver to regenerate rapidly into a fully functional organ has been crucial to the success of living donor liver transplantation (LDLT). In healthy livers, hepatocytes typically remain in a quiescent state (G0). However, following partial hepatectomy, these cells transition to the G1 phase to re-enter the cell cycle. Surgical resection induces various stresses, including physical injury, altered blood flow, and increased metabolic demands. These all trigger the activation and suppression of numerous genes involved in tissue repair, regeneration, and functional recovery. Both coding and noncoding RNAs detectable in the bloodstream during this process provide valuable insights into the gene responses driving liver recovery. This study integrates clinical gene expression data into a previously developed mathematical model of liver regeneration, which tracks transitions among quiescent, primed, and proliferating hepatocytes to construct virtual, patient-specific liver models. Using whole transcriptome RNA sequencing data from 12 healthy LDLT donors, collected at 14 time points over a year, we identified liver resection-specific gene expression patterns through Weighted Gene Co-expression Network Analysis (WGCNA). These patterns were organized into distinct clusters with unique transcriptional dynamics and mapped to model variables using deep learning techniques. Consequently, we developed a Personalized Progressive Mechanistic Digital Twin (PePMDT) for the livers of LDLT donors. The resulting PePMDT predicts individual patient recovery trajectories by leveraging blood-derived gene expression data to simulate regenerative responses. By transforming gene expression profiles into dynamic model variables, this approach bridges clinical data and mathematical modeling, providing a robust platform for personalized medicine. This study highlights the transformative potential of data-driven frameworks like PePMDT in advancing precision medicine and optimizing recovery outcomes for LDLT donors.

Keywords: Deep learning; Digital twin; Liver regeneration; Living donor liver transplantation (LDLT); Mathematical modeling; Partial hepatectomy.

Figure 1:. Framework of the digital twin:

The general framework of the digital twin for liver regeneration. Blood samples are collected from a specific LDLT donor after partial hepatectomy, followed by whole transcriptomic analysis. Identified genes are clustered into functional groups and mapped onto a mechanistic mathematical model describing liver regeneration. The model predicts dynamic variables at future time points, which are then used to forecast future gene expression patterns, enabling a data-driven approach to personalized liver recovery assessment.

Figure 2:. Data structure:

The dataset consists of samples obtained from 12 healthy LDLT patients, collected at 14 distinct time points over the course of one year following partial hepatectomy. Filled color boxes indicate the presence of data, while blank spaces represent missing data. Patients are arranged in descending order based on the number of available time points. The number of available data points out of the 14 total time points is indicated on the right side for each patient. The top two-thirds of the dataset is used for training, while the remaining one-third serves as the testing dataset. The lower portion of the dataset was chosen for testing because these samples lack gene expression values in the later stages, providing an opportunity for predictive modeling.

Figure 3:. Schematic overview of liver regeneration.

(a) The cell cycle dynamics during liver regeneration, highlighting changes in cell numbers. (b) The biochemical pathways involved in liver regeneration, illustrating key molecular interactions driving hepatocyte proliferation and tissue remodeling.

Figure 4:. Schematic representation of the digital twin framework.

The process involves a two-step mapping approach combined with mathematical model simulations. First, gene expression data is mapped onto key physiological parameters of the mathematical model. The model then simulates liver regeneration dynamics over future time points. Finally, the simulated mathematical values are reverse-mapped to reconstruct gene expression values, enabling the prediction of molecular responses in LDLT donors over time.

Figure 5:. Temporal gene modules:

A visual representation of the temporal gene expression patterns across 15 distinct clusters, which were subsequently grouped into 3 modules, during the liver resection and regeneration process. The bold lines represent the mean log2 fold change relative to preoperative values, and the shaded regions indicate the standard deviations.

Figure 6:. Comparison of original and predicted gene expression profiles.

The plot illustrates the temporal gene expression patterns for Patient 5 (a), Patient 6 (b), Patient 10 (c), and Patient 11 (d). Bold lines represent the mean log2 fold change relative to preoperative values, with shaded regions indicating standard deviations. Red lines denote the predicted values, while blue lines represent the original (true) gene expression values. White dots indicate time points where data is missing in the original dataset.

Figure 7:. Patient-specific prediction accuracy.

Comparison of prediction accuracy across individual test patients for the initial time-point (before surgery). Panel A shows the Mean Squared Error (MSE) and Pearson correlation coefficients between the true and predicted gene expression profiles for each cluster. Panel B displays the averaged MSE and averaged Pearson correlation across clusters.

Figure 8:. Comparison of prediction accuracy with different initial time-points (Continued figure: Part 1 for Patient 5).

The figure depicts error bars representing the mean squared errors between the true values and the predicted values for each time-point for patient 5. The red, green, and blue bars correspond to the initial time-points chosen as before surgery, 5 minutes post-surgery, and 30 minutes post-surgery, respectively. This comparison is shown for all 15 clusters.

Figure 8:. Comparison of prediction accuracy with different initial time-points (Continued figure: Part 2 for Patient 6).

The figure depicts error bars representing the mean squared errors between the true values and the predicted values for each time-point for patient 6. The red, green, and blue bars correspond to the initial time-points chosen as before surgery, 5 minutes post-surgery, and 30 minutes post-surgery, respectively. This comparison is shown for all 15 clusters.

Figure 8:. Comparison of prediction accuracy with different initial time-points (Continued figure: Part 3 for Patient 10).

The figure depicts error bars representing the mean squared errors between the true values and the predicted values for each time-point for patient 10. The red, green, and blue bars correspond to the initial time-points chosen as before surgery, 5 minutes post-surgery, and 30 minutes post-surgery, respectively. This comparison is shown for all 15 clusters.

Figure 8:. Comparison of prediction accuracy with different initial time-points (Continued figure: Part 4 (last part) for Patient 11).

The figure depicts error bars representing the mean squared errors between the true values and the predicted values for each time-point for patient 11. The red, green, and blue bars correspond to the initial time-points chosen as before surgery, 5 minutes post-surgery, and 30 minutes post-surgery, respectively. This comparison is shown for all 15 clusters.