An efficient gene regulatory network inference algorithm for early Drosophila melanogaster embryogenesis

Hirotaka Matsumoto, Hisanori Kiryu, Yasuhiro Kojima, Suguru Yaginuma, Itoshi NikaidobioRxiv 2017 · DOI

TL;DR

Researchers have developed a novel, highly efficient algorithm to infer gene regulatory networks (GRNs) from spatial gene expression data. This new method significantly speeds up the process compared to traditional simulation-based approaches, making it practical for analyzing the increasingly large datasets generated by modern experimental techniques.

What Problem Does This Paper Address?

Understanding how genes interact to create complex spatial patterns during embryonic development is a fundamental challenge in developmental biology. In early Drosophila melanogaster (fruit fly) embryos, the precise spatial patterns of gene expression, particularly for "gap genes," are crucial for establishing the body plan. Inferring the underlying gene regulatory network (GRN) that governs these patterns is key to understanding this process. However, existing methods, which often rely on computationally intensive simulations to match observed patterns, are too slow to handle the vast amounts of data now becoming available from advanced imaging and sequencing technologies. This paper aims to bridge this gap by providing an efficient algorithm for GRN inference.

Methods

The core of the paper's contribution lies in a new GRN inference algorithm based on a linear reaction-diffusion (RD) model. While traditional RD models are complex and difficult to solve analytically, the authors simplify this by assuming the system is in a steady state. This allows them to derive a linear ordinary differential equation (ODE) that can be solved analytically. The GRN, represented by a matrix of regulatory interactions, can then be efficiently inferred from the parameters of this ODE using linear regression. This approach bypasses the need for iterative simulations, drastically reducing computation time.

Datasets

The study utilized several datasets:

  • Drosophila melanogaster gap gene expression data: Processed data from SuperFly, covering four trunk gap genes (hb, Kr, gt, kni) along the anterior-posterior (AP) axis.
  • Drosophila melanogaster pair-rule gene expression data: Data for pair-rule genes (h, eve, runt, odd) from the VirtualEmbryos dataset.
  • Clogmia albipunctata and Megaselia abdita gap gene expression data: Data from SuperFly for comparative analysis.
  • Simulated data: Generated from a spatial ODE to qualitatively validate the algorithm's performance.

Key Insights and Ideas

  • Mutual repression is fundamental: The analysis consistently highlights that "mutual repression" (where two genes inhibit each other's expression) is a critical regulatory motif in gap and pair-rule gene networks, essential for forming precise boundaries.
  • Asymmetric regulations fine-tune patterns: Beyond mutual repression, the algorithm identified "asymmetric regulations" that, while not drastically altering peak positions, can subtly adjust them, suggesting a role in fine-tuning spatial patterning.
  • Cross-species GRN comparison is feasible: The algorithm's efficiency and accuracy allow for meaningful comparisons of GRNs between different species, revealing evolutionary differences in regulatory mechanisms (e.g., the distinct GRN in C. albipunctata compared to D. melanogaster).
  • Model selection is integrated: The framework naturally lends itself to model selection, allowing researchers to assess the importance of individual genes within a network.

Key Results

  • Speed: The algorithm achieves GRN inference in sub-second time (around 0.25 seconds for the D. melanogaster gap gene network), a dramatic improvement over simulation-based methods that can take 8 to 160 hours.
  • Accuracy: The inferred GRNs were highly correlated with those obtained from simulation-based approaches (Spearman's rank correlation coefficients of ~0.97 and ~0.91 for comparisons with two different simulation-based GRNs).
  • Qualitative validation: The algorithm successfully identified key regulatory interactions, including mutual repressions, in both simulated and real gene expression data.
  • Species-specific differences: The study revealed that C. albipunctata exhibits a different gap gene network structure, with a more complete mutual repression among three genes (hb, Kr, knl), compared to D. melanogaster.

Limitations and Future Work

  • Linearity assumption: The algorithm relies on a linear reaction-diffusion model and a steady-state assumption. While this enables efficiency, it might not fully capture the complex non-linear dynamics present in biological systems. The authors acknowledge that the optimized model cannot perfectly represent sharp peaks in observed data.
  • Data requirements: The method requires spatially resolved gene expression data, and its performance will depend on the quality and resolution of this data.
  • Future Directions: The authors suggest that the algorithm's efficiency and flexibility make it suitable for analyzing larger GRNs, comparing GRNs across species, and for other analyses like model selection and investigating GRN evolution. The potential for applying this to other developmental systems or even non-biological network inference is also implied.

Summary updated

Abstract

Abstract The spatial patterns of gene expression in early Drosophila melanogaster embryogenesis have been studied experimentally and theoretically to reveal the molecular basis of morphogenesis. In particular, the gene regulatory network (GRN) of gap genes has been investigated through mathematical modeling and simulation. Although these simulation-based approaches are useful for describing complex dynamics and have revealed several important regulations in spatial patterning, they are computationally intensive because they optimize GRN with iterative simulation. Recently, the advance of experimental technologies is enabling the acquisition of comprehensive spatial expression data, and an efficient algorithm will be necessary to analyze such large-scale data. In this research, we developed an efficient algorithm to infer the GRN based on a linear reaction-diffusion model. First, we qualitatively analyzed the GRNs of gap genes and pair-rule genes based on our algorithm and showed that two mutual repressions are fundamental regulations. Then, we inferred the GRN from gap gene data, and identified asymmetric regulations in addition to the two mutual repressions. We analyzed the effect of these asymmetric regulations on spatial patterns, and showed that they have the potential to adjust peak position. Our algorithm runs in sub-second time, which is significantly smaller than the runtime of simulation-based approaches (between 8 and 160 h, for exmaple). Neverthe-less, our inferred GRN was highly correlated with the simulation-based GRNs. We also analyzed the gap gene network of Clogmia albipunctata and showed that different mutual repression regulations might be important in comparison with those of Drosophila melanogaster . As our algorithm can infer GRNs efficiently and can be applied to several different network analysis, it will be a valuable approach for analyzing large-scale data.
BibTeX
@misc{matsumoto2017efficient,
  author = {Hirotaka Matsumoto and Hisanori Kiryu and Yasuhiro Kojima and Suguru Yaginuma and Itoshi Nikaido},
  title = {{An efficient gene regulatory network inference algorithm for early <i>Drosophila melanogaster</i> embryogenesis}},
  year = {2017},
  doi = {10.1101/213025},
  howpublished = {bioRxiv},
}
Figure