Stochastic Chemical Reaction Networks for MAP Detection in Cellular Receivers

Bastian Heinlein, Lukas Brand, Malcolm Egan, Maximilian Schäfer, Robert Schober, Sebastian Lotter

Introduction

In the emerging field of Molecular Communication (MC), information is transmitted via chemical signals rather than electromagnetic waves. This paradigm is particularly promising for biomedical applications where traditional electronics are difficult to deploy. However, a significant bottleneck remains: the design of robust, cellular-scale receivers. Biological systems are inherently stochastic, characterized by random fluctuations in chemical concentrations and reaction timings. While traditional engineering often views this noise as a hindrance, this research explores how the inherent randomness of Chemical Reaction Networks (CRNs) can be leveraged to perform optimal signal detection.

This paper addresses the challenge of building a receiver that is mathematically optimal—specifically achieving Maximum-a-Posteriori (MAP) detection—while remaining practically implementable within the constraints of synthetic biology. The authors propose two architectures that utilize stochasticity as a computational resource rather than a source of error.

Methodology

The researchers developed two distinct CRN architectures designed to perform signal inference in a 3D diffusive environment using Binary Concentration Shift Keying (BCSK).

1. Boltzmann Machine-Inspired CRN

The first approach draws inspiration from machine learning, specifically Boltzmann Machines (BMs). BMs are stochastic neural networks that use Gibbs sampling to reach a thermal equilibrium representing a probability distribution.

To implement this biologically, the authors employed "Taylor Mapping" to translate the mathematical weights and biases of a BM into specific chemical reaction rates. In this system, molecules flip between active (ON) and inactive (OFF) states. The stochastic nature of these transitions allows the network to effectively perform biological Gibbs sampling, calculating the probability of a transmitted signal based on the current state of the cell's receptors.

2. Low-Complexity Adaptive CRN ($C_{LC}$)

Recognizing that a full BM-inspired network might be too complex for certain synthetic biology applications, the authors designed a simplified adaptive CRN. This model introduces a "weight" species ($N_w^{ON}$) that dictates the detection threshold.

This architecture supports online learning through the use of pilot symbols. When the receiver detects a known pilot signal, a feedback mechanism triggers reactions that either increase or decrease the population of weight molecules. This allows the cell to physically "tune" its sensitivity in real-time, adapting to the specific noise characteristics of its environment without requiring an external processor.

Experimental Framework

The study utilized simulated datasets to validate these architectures. The simulation environment consisted of:

  • Channel Model: A 3D diffusive space where molecules move from a transmitter to a spherical receiver.
  • Receiver Model: A cell equipped with 30 to 50 ligand-binding receptors.
  • Modulation: BCSK, where high and low concentrations represent binary 1s and 0s, respectively.
  • Training: The BM-based approach was trained on 10,000 samples to establish baseline weights and biases.

Key Insights and Results

Noise as a Computational Engine

The core insight of this work is that the stochasticity of chemical reactions is the very mechanism that enables probabilistic inference. By mathematically proving that these CRNs can satisfy the "MAP property," the authors demonstrate that a chemical system's steady-state equilibrium can correspond exactly to the most likely transmitted symbol given the observed receptor states.

Performance and Convergence

  • Near-Optimal Detection: The BM-based CRN achieved bit error rates (BER) that were nearly identical to a theoretical MAP detector after only 50 training steps.
  • Adaptive Efficiency: The $C_{LC}$ receiver successfully converged to an optimal detection threshold within 100 to 1,000 pilot symbols.
  • Environmental Tracking: In scenarios where background noise levels shifted suddenly, the adaptive CRN demonstrated the ability to track the new optimal weight values, maintaining reliable communication in non-stationary channels.

Discussion and Future Directions

While the proposed CRNs offer a path toward optimal biological receivers, several challenges remain. The adaptation speed of the $C_{LC}$ model may be insufficient for highly volatile environments, suggesting a need for learning rules with improved noise-averaging or larger step sizes. Furthermore, the current models are limited to binary communication; extending these stochastic networks to support M-ary transmission (multiple signal levels) is a necessary next step.

Ultimately, the transition from theoretical CRNs to physical realization in synthetic or genetically engineered cells represents the next major hurdle. However, by demonstrating that cellular noise can be repurposed for optimal detection, this research provides a foundational framework for the next generation of molecular communication systems.

Summary updated

BibTeX
@inproceedings{heinlein2023stochastic,
  author = {Bastian Heinlein and Lukas Brand and Malcolm Egan and Maximilian Schäfer and Robert Schober and Sebastian Lotter},
  title = {{Stochastic Chemical Reaction Networks for MAP Detection in Cellular Receivers}},
  booktitle = {NANOCOM},
  year = {2023},
  doi = {10.1145/3576781.3608709},
}
Figure