Trajectory Clustering in Multi-layer Networks: Identifying Parkinson's Subtypes.

Sanjukta Krishnagopal

Abstract

Many diseases display heterogeneity in clinical features and their progression, indicative of the existence of disease subtypes. Extracting patterns of disease variable progression for subtypes has tremendous application in medicine, for example, in early prognosis and personalized medical therapy. This work present a novel, data-driven, network-based Trajectory Clustering (TC) algorithm for identifying Parkinson's subtypes based on disease trajectory. Modeling patient-variable interactions as a bipartite network, TC first extracts communities of co-expressing disease variables at different stages of progression. Then, it identifies Parkinson's subtypes by clustering similar patient trajectories that are characterized by severity of disease variables through a multi-layer network. Determination of trajectory similarity accounts for direct overlaps between trajectories as well as second-order similarities, i.e., common overlap with a third set of trajectories. This work clusters trajectories across two types of layers: (a) temporal, and (b) ranges of independent outcome variable (representative of disease severity), both of which yield four distinct subtypes. The former subtypes exhibit differences in progression of disease domains (Cognitive, Mental Health etc.), whereas the latter subtypes exhibit different degrees of progression, i.e., some remain mild, whereas others show significant deterioration after 5 years. The TC approach is validated through statistical analyses and consistency of the identified subtypes with medical literature. This generalizable and robust method can easily be extended to other progressive multi-variate disease datasets, and can effectively assist in targeted subtype-specific treatment in the field of personalized medicine.

Introduction

Parkinson’s Disease (PD) is notoriously heterogeneous; two patients might share a diagnosis but experience vastly different symptoms and rates of progression. Identifying these subtypes is critical for personalized medicine, yet most current methods rely on baseline (one-time) data or lack interpretability. Disease progression is a dynamic process—a "movie" rather than a "still photo." Traditional clustering often fails to capture the complex interplay between clinical variables—such as cognitive decline, sleep issues, and motor impairment—as they evolve over years. This research introduces a data-driven, network-based approach called Trajectory Clustering (TC) that is both longitudinal and transparent.

Methodology

The researchers developed a Stacked Multi-layer Network architecture and a Trajectory Clustering (TC) algorithm. The process follows three main steps:

  1. Bipartite Network Modeling: For every "layer" (either a time-point or a range of disease severity), the algorithm creates a bipartite graph where nodes represent patients and clinical variables. Edges are weighted based on the severity of a patient's symptoms.
  2. Community Detection: Using the Louvain method, the algorithm identifies "variable-communities"—groups of symptoms that tend to co-express with high severity (e.g., depression and anxiety clustering together).
  3. Trajectory Clustering: A patient’s trajectory is defined by their membership in these variable-communities across different layers. The TC algorithm calculates second-order similarity between trajectories, meaning two patients are considered similar if their paths overlap significantly with a common third set of trajectories.
Bipartite Network and Stacked Multi-layer Network architecture for Trajectory Clustering.

The study utilized the Parkinson’s Progression Markers Initiative (PPMI) database, focusing on 194 patients with complete data across five time-points (Baseline to Year 4). Seventeen clinical variables were categorized into domains including Cognitive, Behavioral, Sleep, PD Severity, Autonomic, and Disability.

Key Results and Insights

The TC algorithm identified four distinct subtypes in both temporal and severity-based analyses:

  • Subtype $S_0$ (Mild): Characterized by an older population with relatively mild disease progression, though with specific cognitive impairment (JOLO test).
  • Subtype $S_1$ (Cognitive): Shows consistently high severity across most cognitive variables but remains relatively low in motor disability.
  • Subtype $S_2$ (Motor/Disability): The most severe group, showing rapid deterioration in motor skills, mental health, and daily living activities.
  • Subtype $S_3$ (Autonomic/Sleep): Characterized by severe sleep and autonomic issues, which interestingly appeared to reduce over time in this specific group.
Evolution of variable communities across temporal layers from Baseline to Year 4.

A major advantage of this method is interpretability via Variable-Communities. Unlike standard clustering, TC identifies which symptoms drive the subtype at each stage. For example, mental health variables (GDS, STAI) start as an independent cluster but eventually merge with motor severity and disability clusters as the disease progresses. Furthermore, by using second-order similarity, the algorithm is robust to noise and captures the "neighborhood" of a trajectory, ensuring patients with fundamentally similar progressions are grouped together.

Conclusion

The Trajectory Clustering algorithm provides a flexible and computationally efficient framework for identifying disease subtypes. By stacking layers based on severity scores rather than just time, the model can identify subtypes even when temporal data is inconsistent. While larger datasets are needed to ensure these subtypes are universal, the TC approach is generalizable and can be extended to include genetic markers or bio-fluid biomarkers, offering a significant step forward for personalized medicine in progressive diseases.

Summary updated

Abstract

Many diseases display heterogeneity in clinical features and their progression, indicative of the existence of disease subtypes. Extracting patterns of disease variable progression for subtypes has tremendous application in medicine, for example, in early prognosis and personalized medical therapy. This work present a novel, data-driven, network-based Trajectory Clustering (TC) algorithm for identifying Parkinson's subtypes based on disease trajectory. Modeling patient-variable interactions as a bipartite network, TC first extracts communities of co-expressing disease variables at different stages of progression. Then, it identifies Parkinson's subtypes by clustering similar patient trajectories that are characterized by severity of disease variables through a multi-layer network. Determination of trajectory similarity accounts for direct overlaps between trajectories as well as second-order similarities, i.e., common overlap with a third set of trajectories. This work clusters trajectories across two types of layers: (a) temporal, and (b) ranges of independent outcome variable (representative of disease severity), both of which yield four distinct subtypes. The former subtypes exhibit differences in progression of disease domains (Cognitive, Mental Health etc.), whereas the latter subtypes exhibit different degrees of progression, i.e., some remain mild, whereas others show significant deterioration after 5 years. The TC approach is validated through statistical analyses and consistency of the identified subtypes with medical literature. This generalizable and robust method can easily be extended to other progressive multi-variate disease datasets, and can effectively assist in targeted subtype-specific treatment in the field of personalized medicine.
BibTeX
@misc{krishnagopal2020trajectory,
  author = {Sanjukta Krishnagopal},
  title = {{Trajectory Clustering in Multi-layer Networks: Identifying Parkinson's Subtypes.}},
  year = {2020},
  eprint = {2005.14472},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2005.14472},
}
Figure