Robust predicate and function computation in continuous chemical reaction networks

TL;DR

This paper introduces a new notion of robust computation in chemical reaction networks (CRNs), showing they can reliably decide complex Boolean predicates and compute piecewise affine functions despite rate uncertainties.

Contribution

It defines and analyzes robust computation in CRNs, demonstrating their ability to decide all finite Boolean combinations of threshold predicates and compute piecewise affine functions.

Findings

CRNs can robustly decide all finite Boolean combinations of threshold predicates.

CRNs can robustly compute any piecewise affine function with rational coefficients.

Robust computation is achievable despite rate variations, unlike stable computation.

Abstract

We initiate the study of rate-constant-independent computation of Boolean predicates and numerical functions in the continuous model of chemical reaction networks (CRNs), which model the amount of a chemical species as a nonnegative, real-valued *concentration*. Real-valued numerical functions have previously been studied, finding that exactly the continuous, piecewise rational linear functions $f: \mathbb{R}_{> 0}^k \to \mathbb{R}_{> 0}$ can be computed *stably*, a.k.a., *rate-independently*, meaning that the CRN gets the answer correct no matter the rate at which reactions occur. We show that, contrary to functions, continuous CRNs are severely limited in the Boolean predicates they can stably decide, reporting an answer based only on which inputs are 0 or positive. This limitation motivates a slightly relaxed notion of rate-independent computation in CRNs that we call *robust computation*. The standard mass-action rate model is used, in which each reaction is assigned a rate equal to the product of its reactant concentrations and its rate constant. The computation is correct in this model if it converges to the correct output for any positive choice of rate constants. This adversary is weaker than the stable computation adversary, the latter being able to run reactions at non-mass-action rates. We show that CRNs can robustly decide every finite Boolean combination of *threshold predicates*: those predicates defined by taking a rational weighted sum of the inputs $\mathbf{x} \in \mathbb{R}^k_{\ge 0}$ and comparing to a constant, answering the question ``Is $\sum_{i=1}^k w_i \cdot \mathbf{x}(i) > h$?'', for rational weights $w_i$ and real threshold $h$. Turning to function computation, we show that CRNs can robustly compute any piecewise affine function with rational coefficients, where threshold predicates determine which affine piece to evaluate for a given input.