Improved supervised prediction of aging-related genes via weighted dynamic network analysis

Qi Li, Khalique Newaz, Tijana Milenković

Abstract

Predicting aging-related genes is a critical task in understanding biological decline. While gene expression data captures temporal dynamics and protein-protein interaction (PPI) networks capture functional relationships, integrating them effectively remains a challenge. Previous "Induced" dynamic subnetworks often failed to outperform static models. This study proposes a weighted dynamic subnetwork using Network Propagation (NetWalk). We demonstrate that accounting for aging-specific edge weights in a dynamic context significantly improves prediction accuracy (AUPR and F-score) over existing unweighted and static methods.

Introduction

Aging is inherently dynamic, yet many computational models rely on static PPI networks that ignore temporal changes in protein interactions. While dynamic subnetworks can be constructed by integrating gene expression data, early methods used a naive "Induced" approach—simply including an edge if both genes are active—which treats all interactions as equally important. This study argues that the "secret sauce" for accurate prediction lies in capturing the varying strengths, or weights, of these interactions as they evolve over time.

Methodology

The researchers moved beyond the simple on/off logic of the Induced approach by employing Network Propagation (NP), specifically the NetWalk algorithm. NP allows for the identification of "bridge" genes that may not show high expression changes but are topologically central to the aging process.

The study compared several network architectures to isolate the impact of dynamic information and edge weighting.

Classification of the network models evaluated in the study, distinguishing between static/dynamic and weighted/unweighted approaches.

The proposed wNetWalk-Dynamic model utilizes continuous interaction weights. To capture the progression of aging, the authors created "differential snapshots" measuring the percentage change in weights between consecutive ages (20 to 99 years). They also introduced the Diff-nobin-2 feature set to quantify a gene's evolving role within these weighted structures.

Results and Discussion

The models were evaluated on their ability to predict genes in the GenAge and GTEx-DAG databases. The results clearly indicate that the proposed weighted dynamic approach is superior.

Performance metrics (AUPR, Precision, Recall, and F-score) for the various models. The 'wNetWalk-Dynamic Proposed' model consistently achieves the highest accuracy.

Key insights from the evaluation include:

  • Weighting is Essential: Unweighted dynamic models often performed worse than static ones. Only when aging-specific weights were added did the dynamic models consistently outperform the static baselines.
  • NP vs. Induced: Network Propagation (NetWalk) proved more effective than the Induced approach at capturing the underlying biology of aging.
  • Biological Validation: Novel gene predictions (those not in the GenAge "ground truth") showed a 34% overlap with cancer-related datasets, reinforcing the known biological link between aging and oncogenesis. Top candidates such as ATF1, GLI3, and SMC1A were further validated through literature searches linking them to age-related pathologies.

Conclusion

This research highlights the necessity of using weighted dynamic models to study the aging process. By integrating temporal gene expression with PPI networks through weighted network propagation, the proposed method achieves state-of-the-art performance in aging-related gene prediction. Future research will focus on incorporating more comprehensive interactome data and exploring deep learning architectures like Graph Convolutional Networks.

Summary updated

Abstract

This study focuses on the task of supervised prediction of aging-related genes from -omics data. Unlike gene expression methods for this task that capture aging-specific information but ignore interactions between genes (i.e., their protein products), or protein-protein interaction (PPI) network methods for this task that account for PPIs but the PPIs are context-unspecific, we recently integrated the two data types into an aging-specific PPI subnetwork, which yielded more accurate aging-related gene predictions. However, a dynamic aging-specific subnetwork did not improve prediction performance compared to a static aging-specific subnetwork, despite the aging process being dynamic. This could be because the dynamic subnetwork was inferred using a naive Induced subgraph approach. Instead, we recently inferred a dynamic aging-specific subnetwork using a methodologically more advanced notion of network propagation (NP), which improved upon Induced dynamic aging-specific subnetwork in a different task, that of unsupervised analyses of the aging process. Here, we evaluate whether our existing NP-based dynamic subnetwork will improve upon the dynamic as well as static subnetwork constructed by the Induced approach in the considered task of supervised prediction of aging-related genes. The existing NP-based subnetwork is unweighted, i.e., it gives equal importance to each of the aging-specific PPIs. Because accounting for aging-specific edge weights might be important, we additionally propose a weighted NP-based dynamic aging-specific subnetwork. We demonstrate that a predictive machine learning model trained and tested on the weighted subnetwork yields higher accuracy when predicting aging-related genes than predictive models run on the existing unweighted dynamic or static subnetworks, regardless of whether the existing subnetworks were inferred using NP or the Induced approach.
BibTeX
@misc{li2020improved,
  author = {Qi Li and Khalique Newaz and Tijana Milenković},
  title = {{Improved supervised prediction of aging-related genes via weighted dynamic network analysis}},
  year = {2020},
  eprint = {2005.03659},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2005.03659},
}
Figure