Maximizing Output and Recognizing Autocatalysis in Chemical Reaction Networks is NP-Complete

Background: A classical problem in metabolic design is to maximize the production of desired compound in a given chemical reaction network by appropriately directing the mass flow through the network. Computationally, this problem is addressed as a linear optimization problem over the “flux cone”. The prior construction of the flux cone is computationally expensive and no polynomial-time algorithms are known. Results: Here we show that the output maximization problem in chemical reaction networks is NP-complete. This statement remains true even if all reactions are monomolecular or bimolecular and if only a single molecular species is used as influx. As a corollary we show, furthermore, that the detection of autocatalytic species, i.e., types that can only be produced from the influx material when they are present in the initial reaction mixture, is an NP-complete computational problem. Conclusions: Hardness results on combinatorial problems and optimization problems are important to guide the development of computational tools for the analysis of metabolic networks in particular and chemical reaction networks in general. Our results indicate that efficient heuristics and approximate algorithms need to be employed for the analysis of large chemical networks since even conceptually simple flow problems are provably intractable.

💡 Research Summary

The paper investigates two fundamental computational problems in the analysis of chemical reaction networks (CRNs): (i) maximizing the production of a desired compound (output maximization) and (ii) detecting autocatalytic species, i.e., compounds that can only be generated from the influx material when they are already present in the reaction mixture. The authors formalize CRNs as directed multi‑hypergraphs where vertices represent chemical species and hyper‑edges encode reactions with multisets of reactants and products. A flow f assigns a non‑negative integer to each hyper‑edge and must satisfy a mass‑balance condition at every internal vertex, guaranteeing a stationary state. External reservoirs are modeled by pseudo‑reactions e_in(x) and e_out(x) that inject or extract material.

Three decision/optimization variants are defined: MAX‑CRN‑Output (general case), MAX‑CRN(d)‑Output (each reaction has in‑degree and out‑degree ≤ d), and MAX‑CRN(d)‑Output‑1 (only a single influx vertex). The autocatalysis decision problem asks whether there exists a species x such that (1) it cannot be produced without its own input (any flow with f(e_in(x)) = 0 forces f(e_out(x)) = 0) and (2) there exists a flow where the output of x exceeds its input, i.e., the species can amplify itself.

To prove NP‑completeness, the authors reduce the strongly NP‑complete 3‑Partition problem to each CRN variant. The reduction proceeds in two stages. First, they construct an intermediate lattice‑like network (Figure 2A) consisting of input nodes (one per integer s_i), auxiliary nodes Z_j that receive a fractional share of the total sum, a grid of “switch” nodes, waste nodes, and a single output sink O. Each switch can be in an “off” state (passing left input to the right and top input downwards) or an “on” state (consuming its left input, using an equal amount of top input, and diverting that amount to the output). By assigning exactly three “on” switches per column and one per row, a feasible assignment corresponds precisely to a partition of the s_i into m triples of equal sum. The total flow reaching O equals the sum of all s_i if and only if the 3‑Partition instance is solvable; otherwise some flow is lost to waste.

In the second stage, each abstract switch is replaced by a concrete sub‑CRN built from monomolecular or bimolecular reactions. The sub‑CRN uses copies of Q_i (representing the integer s_i) and Z_j (representing the target sum per subset) to produce s_i copies of an output molecule O only when the required numbers of Q_i and Z_j are present. Additional “drain” reactions route excess material to waste. This construction preserves the property that the maximal achievable output flow equals the total sum of the s_i precisely when a valid 3‑Partition exists. Moreover, the construction can be adapted so that all reactions are at most bimolecular and the network has only a single influx species, establishing NP‑completeness for MAX‑CRN(d)‑Output‑1.

For the autocatalysis problem, the authors modify the same construction: they designate a particular species x as both an input and an output node. The network is arranged such that x can only be produced in excess of its influx if the underlying 3‑Partition instance has a solution; otherwise any flow that produces x must consume at least as much x as it injects, violating the autocatalysis condition. Hence detecting an autocatalytic species is also NP‑complete.

The paper’s contributions are threefold: (1) it proves that output maximization in CRNs is NP‑complete even under severe restrictions (monomolecular/bimolecular reactions, single input), (2) it shows that identifying autocatalytic compounds is NP‑complete, and (3) it highlights that, despite the polynomial‑time solvability of related problems such as finding a single elementary flux mode, the optimization problems central to metabolic engineering are computationally intractable in the worst case. Consequently, the authors argue that practical analysis of large metabolic or chemical networks must rely on heuristics, approximation algorithms, or problem‑specific tractable subclasses rather than exact optimization.