DNFE: Directed-network flow entropy for detecting the tipping points during biological processes

Xueqing Peng, Peiluan Li, Luonan ChenbioRxiv 2024 · DOI

TL;DR

This paper introduces Directed-network Flow Entropy (DNFE), a novel computational method to identify critical transition points in biological processes using directed gene networks. DNFE is more robust and effective than traditional methods, especially for high-dimensional single-cell data, and can uncover "dark genes" and transcription factors crucial for understanding disease progression and cell fate decisions.

What Problem Does This Paper Address?

Biological systems, from cell differentiation to disease progression, often undergo abrupt "tipping points" where a system shifts from one stable state to another. Identifying these critical transition points is vital for understanding underlying mechanisms and potentially intervening to prevent catastrophic outcomes. However, existing methods often struggle with the complexity and high dimensionality of biological data, particularly single-cell RNA sequencing (scRNA-seq) data, which is prone to noise and dropout events. Traditional approaches typically rely on undirected networks, failing to capture the crucial directional relationships between genes that drive these transitions. This limits their effectiveness in pinpointing tipping points and identifying key regulatory factors.

Methods

The core innovation is the Directed-network Flow Entropy (DNFE) method. It transforms gene expression data into a directed network, where edges represent gene interactions with a defined direction. This is achieved by:

  1. Constructing a directed network: This involves rewiring a weighted gene coexpression network (WGCNA) using a direction determination index that leverages gene expression profiles and sample phenotypes.
  2. Extracting local gene modules: For each gene, a local network is built considering its first and second-order neighbors.
  3. Calculating local DNFE: A measure of network flow entropy is computed for each local directed network. Unlike traditional entropy, DNFE specifically quantifies network fluctuation by utilizing local network connectivity information, making it less susceptible to noise from individual gene expression levels.
  4. Calculating global DNFE: The local DNFE scores are aggregated to obtain a global DNFE score for the entire system.

A significant increase in the global DNFE score over time serves as an early warning signal for an impending tipping point. The method also identifies Dynamic Network Biomarkers (DNBs) – genes with high DNFE scores that are sensitive to critical states – and "dark genes" – genes that are not differentially expressed but show significant network fluctuation.

Datasets

The DNFE method was validated on five real-world datasets:

  • Three single-cell RNA-sequencing (scRNA-seq) datasets:
    • hESC-to-DEC (human embryonic stem cells to definitive endoderm)
    • mESC-to-MP (mouse embryonic stem cells to mesoderm progenitor)
    • MEF-to-Neurons (mouse embryonic fibroblast to neuron)
  • Two bulk tumor datasets:
    • Kidney renal papillary cell carcinoma (KIRP)
    • Bladder urothelial carcinoma (BLCA)

The paper also mentions numerical simulations using artificial gene regulatory networks of 11, 100, and 1000 nodes to assess robustness under varying noise strengths.

Key Insights and Ideas

  • Directed networks are crucial: The paper strongly advocates for the use of directed networks over undirected ones to accurately capture gene regulatory relationships and identify tipping points.
  • DNFE as an early warning system: The DNFE score's ability to show a sharp increase as a system approaches a critical state makes it a powerful tool for early detection and prediction.
  • Uncovering "dark genes": DNFE can identify genes that are not differentially expressed but are critical to network dynamics, offering a new perspective on biological regulation and disease mechanisms. These "dark genes" can be overlooked by traditional expression-based analyses.
  • Robustness and effectiveness: DNFE demonstrates superior performance and robustness compared to existing methods, especially in handling noisy and high-dimensional scRNA-seq data.
  • Predicting transcription factors: The method can identify key transcription factors (TFs) that play pivotal roles in cell fate determination and disease progression.

Key Results

  • DNFE successfully detected critical states and tipping points in all five real datasets (three scRNA-seq, two bulk tumor).
  • Numerical simulations showed DNFE to be robust and effective across different noise levels and network sizes (up to 1000 nodes).
  • DNFE outperformed the sJSD method in providing early-warning signals under varying noise strengths.
  • In the hESC-to-DEC differentiation, DNFE identified two tipping points at 12h and 36h, correlating with distinct differentiation processes.
  • The method identified key transcription factors involved in hESC differentiation and predicted ZNF888 as a potentially important, yet previously overlooked, biomarker.
  • For KIRP and BLCA tumor datasets, DNFE identified critical stages and provided prognostic insights, showing that DNFE-based stratification of patients was more effective than gene expression-based stratification for predicting survival.
  • DNFE identified "dark genes" that were significantly enriched in cancer-related pathways like JAK-STAT, PI3K-Akt, and cell cycle, suggesting their crucial role in disease progression.

Limitations and Future Work

  • Experimental Validation: While DNFE identifies potential biomarkers and regulatory mechanisms, further experimental validation is explicitly stated as necessary, particularly for the identified transcription factors like ZNF888.
  • Parameter Tuning: The paper mentions that the value of lambda (λ) in the local directed network calculation can be determined on a case-by-case basis, implying that optimal parameter selection might be required for different datasets.
  • Complexity of Biological Systems: The paper focuses on gene-level interactions. Future work could explore extending DNFE to incorporate other molecular layers (e.g., proteins, metabolites) for a more comprehensive understanding of complex biological systems.
  • Broader Applications: The authors suggest that DNFE has promising potential for exploring molecular mechanisms of disease progression and discovering new network biomarkers, indicating a broad scope for future research.

Summary updated

Abstract

ABSTRACT There generally exists a critical state or tipping point from a stable state to another in dynamic biological processes, beyond which a significant qualitative transition occurs. Identifying this tipping point and its driving network is essential to prevent or delay catastrophic consequences. However, most traditional approaches based on undirected networks still suffer from the problem of the robustness and effectiveness when applied to high-dimensional small sample data, especially for single-cell data. To address this challenge, we developed a directed-network flow entropy (DNFE) method which can transform measured omics data into a directed network. This method is applicable to both single-cell RNA-sequencing (scRNA-seq) and bulk data. By applying this method to five real datasets, including three single-cell datasets and two bulk tumor datasets, the method can not only successfully detect the critical states as well as their dynamic network biomarkers, but also help explore regulatory relationships between genes. Numerical simulation indicates that the DNFE method is robust and superior to existing methods. Furthermore, DNFE has predicted active transcription factors (TFs), and further identified ‘dark genes’, which are usually overlooked by traditional methods.
BibTeX
@misc{peng2024dnfe,
  author = {Xueqing Peng and Peiluan Li and Luonan Chen},
  title = {{DNFE: Directed-network flow entropy for detecting the tipping points during biological processes}},
  year = {2024},
  doi = {10.1101/2024.09.18.613673},
  howpublished = {bioRxiv},
}
Figure