Sequential Reprogramming of Boolean Networks Made Practical

Hugues Mandon, Cui Su, Stefan Haar, Jun Pang, Loïc PaulevéCMSB 2019 · DOI

TL;DR

This paper introduces a new method for "reprogramming" gene regulatory networks, which are modeled as Boolean networks. Their approach, called attractor-based sequential reprogramming, allows for more practical and biologically feasible control of these complex systems by only considering stable states (attractors) as intermediate steps, unlike previous methods that required complete system observability.

What Problem Does This Paper Address?

Cell reprogramming, the process of converting one cell type into another, is a cornerstone of regenerative medicine. Identifying the right genetic "switches" (transcription factors) to achieve this conversion is a major challenge, as blindly testing combinations is prohibitively expensive. Computational models, particularly Boolean networks, offer a way to predict these genetic interventions. However, existing computational methods for reprogramming often require either instantaneous, large-scale perturbations (one-step reprogramming) or precise knowledge of the system's state at all times (sequential reprogramming), which is rarely achievable in real biological experiments. This paper addresses the need for a reprogramming strategy that is both computationally efficient and practically applicable in a biological context, by leveraging the inherent stability of biological systems.

Methods

The core of this paper is the development of an attractor-based sequential reprogramming algorithm. This method builds upon the concept of Boolean networks and their dynamics, which are characterized by attractors – stable states or cycles that the network eventually settles into.

Here's a breakdown of their approach:

  • Attractors as Intermediates: Instead of requiring the system to be in any arbitrary state for reprogramming, their method only uses attractors as intermediate points. This is crucial because attractors represent observable, stable phenotypes of the cell.
  • Sequential Perturbations: The reprogramming is achieved through a sequence of perturbations, where each perturbation aims to guide the network from one attractor to another (either an intermediate attractor or the final target attractor).
  • Modified Hamming Distance: To account for practical constraints, they introduce a modified Hamming distance that penalizes perturbations to certain "forbidden" genes or nodes.
  • Algorithm: They propose an algorithm that systematically searches for sequences of attractors and corresponding minimal perturbations to reach a target attractor from a source attractor, while respecting constraints like avoiding specific intermediate attractors or perturbing certain nodes.

Datasets

The authors evaluated their method on several biological networks:

  • Myeloid differentiation network
  • Cardiac gene regulatory network
  • ERBB receptor regulated G1/S transition network
  • Tumour network
  • PC12 cell network
  • Hematopoietic cell specification network
  • Bortezomib responses network

These networks are described as having single-state attractors. The paper does not introduce new benchmark datasets but uses existing biological network models.

Key Insights and Ideas

  • Bridging the Gap: The attractor-based approach offers a practical middle ground between the overly simplistic one-step reprogramming and the overly demanding fully observable sequential reprogramming.
  • Biological Realism: By focusing on attractors as intermediate states, the method aligns better with experimental realities where stable cell states are observable and controllable.
  • Flexibility through Constraints: The ability to specify forbidden nodes for perturbation and avoid certain attractors (like apoptosis) makes the method highly adaptable to specific biological contexts.
  • Reduced Experimental Cost: By identifying sequences of perturbations that leverage natural network dynamics, the method promises to reduce the number of experiments needed to achieve a desired cell state.

Key Results

The proposed attractor-based sequential reprogramming method was compared against minimal one-step reprogramming and minimal sequential reprogramming on several biological networks.

  • Reduced Perturbations: The sequential strategies (both theirs and the minimal sequential one) generally required fewer perturbations than the one-step method, with reductions of up to 9 perturbations observed.
  • Comparable to Minimal Sequential: Their method showed comparable results to the minimal sequential reprogramming in terms of the number of perturbations, but with the significant advantage of using observable attractors as intermediates.
  • Scalability: The implementation is scalable, handling networks with over 60 nodes.
  • Computational Efficiency: All three methods were found to be efficient in terms of computation time, with the attractor-based method taking slightly longer but providing more biologically relevant solutions.
  • Path Diversity: The method can identify multiple sequential paths, some of which may require fewer perturbations than the shortest path, offering a richer set of potential experimental strategies.

Limitations and Future Work

  • Observability of Cyclic Attractors: While the method generally reduces the need for complete observability, it still requires observation of the system's state when intermediate attractors are cyclic.
  • Temporary Perturbations: The current work focuses on instantaneous perturbations. Future work aims to extend the method to handle temporary perturbations, which is another common experimental scenario.
  • Relaxing Observability: Further research is planned to relax the observability requirements even for cyclic attractors.

Summary updated

BibTeX
@inproceedings{mandon2019sequential,
  author = {Hugues Mandon and Cui Su and Stefan Haar and Jun Pang and Loïc Paulevé},
  title = {{Sequential Reprogramming of Boolean Networks Made Practical}},
  booktitle = {CMSB},
  year = {2019},
  doi = {10.1007/978-3-030-31304-3\_1},
}

Loading related content...

Figure