Numerical Reliability of Logistic Gene Regulatory Network Models: Preventing Expression Shutdown and Robust Integration of Boolean-Derived ODE Systems

Ismail Belgacem

arxiv: 2605.01056 · v2 · pith:Q6KKKFK7new · submitted 2026-05-01 · 🧬 q-bio.MN · math.DS

Numerical Reliability of Logistic Gene Regulatory Network Models: Preventing Expression Shutdown and Robust Integration of Boolean-Derived ODE Systems

Ismail Belgacem This is my paper

Pith reviewed 2026-05-09 14:34 UTC · model grok-4.3

classification 🧬 q-bio.MN math.DS

keywords gene regulatory networkslogistic functionsHill functionsnumerical stabilitybistabilitynegative feedback oscillatorBoolean networksgene expression

0 comments

The pith

Logistic functions fix the smoothness, numerical stability, and zero-basal-rate flaws of Hill functions in gene regulatory network models.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that logistic activation functions provide a globally smooth, real-valued, and strictly positive-at-zero replacement for Hill functions when modeling gene regulation. These properties eliminate three problems that arise with non-integer cooperativity exponents: loss of global differentiability, silent insertion of complex numbers into ODE trajectories, and complete shutdown of basal production that traps bistable circuits in inactive states. A reader would care because the change enables provable stability in oscillators, finite escape times from off-states in positive autoregulation, and reliable long-term integration of Boolean-derived networks without solver warnings or failures. Demonstrations cover local stability via Routh-Hurwitz, explicit saddle-node thresholds, automatic De Morgan translation of an 11-gene cell-cycle network, and stable runs to t=200 on an 80-gene system.

Core claim

Logistic functions f±, defined to be globally C^∞, real for every real input, and strictly positive at zero, resolve the three structural flaws of Hill functions simultaneously. In a two-gene negative-feedback oscillator, the Routh-Hurwitz criterion shows local asymptotic stability for all positive parameter values and rules out Hopf bifurcation without time delays. For bistable positive autoregulation, saddle-node thresholds are given by explicit transcendental equations; with E. coli parameters the logistic basal rate produces off-state escape in approximately 44 minutes while the Hill model remains trapped indefinitely. The product-of-logistics De Morgan formalism translates Boolean logic

What carries the argument

The logistic activation functions f± that replace Hill functions, guaranteeing global smoothness, real arithmetic, and positive basal production rate.

Load-bearing premise

That logistic functions with positive basal production represent the underlying biology at least as accurately as Hill functions for the gene circuits and parameter ranges considered.

What would settle it

A laboratory measurement showing that a bistable positive-autoregulation circuit remains permanently off under conditions where the logistic model predicts escape within one hour, or a simulation run in which a Hill model with non-integer exponent produces no complex values or warnings.

Figures

Figures reproduced from arXiv: 2605.01056 by Ismail Belgacem.

**Figure 1.** Figure 1: Architecture of a two-gene negative feedback loop. Gene A activates gene B (blue arrow), while gene B represses gene A (red bar) view at source ↗

**Figure 2.** Figure 2: Temporal evolution of the two-gene oscillator system (4). Parameters: 𝜆 = 3, 𝜅1 = 3, 𝛾1 = 0.25, 𝜅2 = 4, 𝛾2 = 0.5, 𝜃1 = 4, 𝜃2 = 3; initial conditions 𝑥01 = 𝑥02 = 1 view at source ↗

**Figure 3.** Figure 3: Temporal evolution of the 11-gene Traynard cell-cycle logistic ODE system over 𝑡 ∈ [0, 60], starting from the initial conditions in view at source ↗

**Figure 4.** Figure 4: Comparison of the decreasing logistic functions 𝑓 − 1 (𝑥, 𝜃, 𝜆) = 1∕(1 + 𝑒 −𝜆(𝑤𝑥+𝜃) ) (our model) and 𝑓 − 2 (𝑥, 𝜃, 𝜇) = 1∕(1+exp(−𝜇(𝑤𝑥−𝜃))) (Samuilik) with 𝑤 = −1, alongside the decreasing Hill function ℎ − (𝑥, 𝜃, 𝑛) = 𝜃 𝑛∕(𝑥 𝑛+𝜃 𝑛 ). Parameters: 𝑛 = 4, 𝜃 = 3, 𝜆 = 𝑛∕𝜃 ≈ 1.333, 𝜇 = 𝜆. The critical point of 𝑓 − 1 is 𝑥𝑐 = −𝜃∕𝑤 = 3 > 0 (biologically meaningful), while that of 𝑓 − 2 is 𝑥𝑐 = 𝜃∕𝑤 = −3 < 0 (outsid… view at source ↗

**Figure 5.** Figure 5: Trajectories of the genetic oscillator under small perturbations and strong degradation. The logistic model (solid) escapes the low-expression trap, while the Hill model (dashed) stagnates. Left: 𝑥1 (𝑡); Right: 𝑥2 (𝑡). First Author et al.: Page 39 of 37 view at source ↗

**Figure 6.** Figure 6: Protein dynamics in positive autoregulation under low initial conditions (𝑚(0) = 0.01, 𝑥(0) = 0.01) with feedback amplification 𝛼 ≈ 600. Logistic model (monostable-high regime): The system escapes the off-state in approximately 2650 s (∼44 min) due to basal production 𝑘𝑚𝑓(0) ≈ 0.000142 s −1, reaching 𝑥 ≈ 38 at 10,000 s and approaching 𝑥ss ≈ 600 asymptotically. Hill model (bistable regime): mRNA decays expo… view at source ↗

**Figure 7.** Figure 7: Hill-function ODE system (𝑛 ≈ 3.51, non-integer), full time horizon 𝑡 ∈ [0, 200]. Trajectories of all 80 state variables. The solver’s reliable domain ends between 𝑡 = 50 and 𝑡 = 100: beyond that point several curves diverge catastrophically. The 𝑦-axis spans [−500 000, 2 × 106 ]; the most divergent variable (blue curve, 𝑥2 ) reaches ≈ 2 × 106 at 𝑡 = 200, while the most negative variable descends to approx… view at source ↗

**Figure 8.** Figure 8: Hill-function ODE system (𝑛 ≈ 3.51, non-integer), early window 𝑡 ∈ [0, 65]. The same simulation as view at source ↗

**Figure 9.** Figure 9: Logistic-function ODE system (same 𝑛 and 𝜃 parameters), full time horizon 𝑡 ∈ [0, 200]. Trajectories of all 80 state variables. NDSolve completes the integration without any warnings. All variables remain strictly non-negative throughout and converge to bounded steady states; the 𝑦-axis is confined to [0, 275], consistent with the biological bound 𝜅𝑖∕𝛾𝑖 . Most variables settle before 𝑡 = 50. Two variables … view at source ↗

read the original abstract

Gene regulatory networks are routinely translated from Boolean update rules into large continuous ODE systems integrated numerically for attractor identification, sensitivity analysis, and control design. The reliability of that integration depends critically on the sigmoidal kernel representing regulation. This simulation study shows that the Hill function -- the near-universal choice -- is a generically unreliable kernel, while the logistic function is a robust replacement. Two failure modes are demonstrated. First, because the Hill function vanishes at zero input, bistable circuits acquire an absorbing off-state: with experimentally grounded \textit{E. coli} galactose-operon autoregulation parameters, a Hill model stays trapped below the unstable separatrix, whereas the logistic model -- whose basal rate is strictly positive by construction -- escapes in about $44$~minutes through basal production alone, matching an analytical estimate of ${\approx}58$~min. A saddle-node analysis characterises the bistable window via an explicit transcendental equation and identifies the threshold $\lambda\theta=2$ separating monostable from bistable regimes. Second, when the Hill exponent is non-integer -- as in dose-response fits -- the power law $x^n=e^{n\ln x}$ turns complex-valued whenever a solver overshoots into negative concentrations. On an $80$-gene Boolean-derived benchmark with $n\approx3.509$, the Hill solver is silently contaminated by complex values from $t\approx52.64$, yielding smooth but spurious trajectories, whereas the logistic formulation completes $t\in[0,200]$ without a single warning. Because the logistic vector field is globally Lipschitz with explicit constant, we further prove an a priori global-error bound of classical order -- a guarantee structurally unavailable to the Hill formulation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Logistic functions fix numerical headaches in GRN ODEs but trade one modeling choice for another on basal expression.

read the letter

The paper's core move is replacing Hill functions with logistic ones in gene regulatory network ODEs. This keeps everything smooth and real-valued while adding a small positive basal production rate at zero input. The author shows this sidesteps three documented problems: non-integer cooperativity breaking global differentiability, silent complex-value creep in solvers, and permanent trapping in off-states for bistable positive feedback loops. Concrete checks include Routh-Hurwitz stability for a two-gene oscillator (no Hopf without delay), explicit saddle-node conditions for positive autoregulation, and an 80-gene Boolean-derived system that runs cleanly to t=200 under logistic but hits complex contamination under Hill around t=52. The De Morgan product form for repressor logic also keeps thresholds at positive concentrations, unlike some weighted-sum alternatives. These are useful, reproducible numerical results for anyone who has watched a large GRN simulation die on arithmetic warnings. The E. coli bistable example reports off-state escape in roughly 44 minutes under grounded parameters, which follows directly from the logistic property f(0)>0. That is the part worth pausing on. The escape is not an extra feature; it is the direct consequence of replacing a zero basal rate with a positive one. If the real promoters in question have basal transcription consistent with zero within experimental error, the logistic version is changing the model rather than repairing it. The paper grounds parameters biophysically but does not show side-by-side comparison against measured escape times or basal rates, so the claim that this constitutes prevention rather than distortion rests on that modeling choice. The 11-gene cell-cycle translation and the large-network run are still solid evidence that the formulation integrates without warnings. This work is aimed at computational systems biologists who simulate large or stiff GRNs and have hit the Hill-function limitations in practice. The numerical demonstrations and small-system proofs are clear enough that a serious editor should send it to referees, mainly to check the parameter grounding and to ask whether the logistic leak needs explicit validation against data before it is adopted as a default replacement.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes logistic functions f^± as replacements for Hill functions in gene regulatory network (GRN) models. It argues that logistics simultaneously resolve three Hill-function flaws for non-integer cooperativity: loss of global C^∞ smoothness, silent complex-valued arithmetic corruption of ODE trajectories, and identically zero basal production that traps bistable models in off-states. The paper establishes local asymptotic stability of a two-gene negative-feedback oscillator for all positive parameters via the Routh-Hurwitz criterion with no Hopf bifurcation possible without delays; derives saddle-node thresholds for bistable positive autoregulation via explicit transcendental equations; reports off-state escape in ≈44 min under biophysically grounded E. coli parameters; translates an 11-gene Traynard cell-cycle Boolean network via product-of-logistics De Morgan formalism; and demonstrates that an 80-gene Boolean-derived ODE system (n=3.509) completes t∈[0,200] without warnings while the Hill formulation enters complex-valued contamination at t≈52.64.

Significance. If the logistic basal rate is biologically appropriate, the work supplies a globally smooth, real-valued, and fully controllable framework for GRN modeling that avoids numerical artifacts and enables control-theoretic methods. The Routh-Hurwitz proof, explicit transcendental saddle-node conditions, and reproducible numerical demonstrations on both small and 80-gene systems constitute clear strengths. The De Morgan threshold placement at positive concentrations is a useful contrast to prior weighted-sum formulations. Significance is tempered by the need for stronger justification that non-zero basal production improves rather than distorts bistable dynamics.

major comments (2)

[Bistable positive autoregulation] Bistable positive autoregulation section: the claim that logistic f^± 'prevents expression shutdown' rests on the finite mean escape time produced exactly by f(0)>0. The manuscript reports ≈44 min escape under 'biophysically grounded E. coli parameters' but does not list the explicit basal production rate, its literature source, or sensitivity analysis; without these, it is impossible to assess whether the result reflects measured promoter leakiness or post-hoc parameter selection that artifactually destabilizes the off-state.
[Two-gene negative-feedback oscillator] Oscillator stability section: the Routh-Hurwitz criterion is applied to conclude local asymptotic stability for all positive parameters and no Hopf bifurcation without delays. The characteristic equation and the explicit Routh array entries should be provided so readers can verify the sign conditions independently; their absence makes the 'for all positive parameters' claim difficult to check.

minor comments (2)

[Introduction / Methods] The explicit functional forms of the logistic activation f^+ and repression f^- (including the precise definition of the cooperativity parameter) are referenced but not displayed in the abstract or early sections; they should appear with equation numbers before any stability or bifurcation analysis.
[Large-network numerical integration] In the 80-gene numerical example the cooperativity n=3.509 is stated without indicating whether it arises from a fit to data or is chosen for illustration; this detail affects reproducibility of the complex-value failure time t≈52.64.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below and will revise the manuscript to incorporate the requested clarifications and supporting material.

read point-by-point responses

Referee: [Bistable positive autoregulation] Bistable positive autoregulation section: the claim that logistic f^± 'prevents expression shutdown' rests on the finite mean escape time produced exactly by f(0)>0. The manuscript reports ≈44 min escape under 'biophysically grounded E. coli parameters' but does not list the explicit basal production rate, its literature source, or sensitivity analysis; without these, it is impossible to assess whether the result reflects measured promoter leakiness or post-hoc parameter selection that artifactually destabilizes the off-state.

Authors: We agree that explicit documentation is required. In the revised manuscript we will state the precise basal production rate employed for the E. coli parameter set, cite the primary literature sources for measured promoter leakiness in E. coli, and add a sensitivity analysis that varies the basal rate over the range of experimentally reported values. This will show that off-state escape times remain on the order of tens of minutes for biophysically plausible leakiness levels. revision: yes
Referee: [Two-gene negative-feedback oscillator] Oscillator stability section: the Routh-Hurwitz criterion is applied to conclude local asymptotic stability for all positive parameters and no Hopf bifurcation without delays. The characteristic equation and the explicit Routh array entries should be provided so readers can verify the sign conditions independently; their absence makes the 'for all positive parameters' claim difficult to check.

Authors: We will include the full derivation in the revision. The updated section will present the Jacobian of the two-gene system, the resulting characteristic polynomial, and the complete Routh array with all entries and sign conditions. This will allow direct verification that the Hurwitz criteria are satisfied for every positive parameter combination. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivations are self-contained

full rationale

The paper derives local stability for the negative-feedback oscillator directly from the Routh-Hurwitz criterion applied to the logistic ODE system, obtains saddle-node thresholds for positive autoregulation via explicit transcendental equations, and performs direct numerical integration on the 11-gene and 80-gene networks. All results follow from the stated functional forms of f^±, the De Morgan translation rule, and externally supplied biophysical parameters without any fitting of outputs to inputs, self-referential definitions, or load-bearing self-citations. The contrast with Hill functions is presented as a transparent consequence of the basal-rate difference rather than a derived prediction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard properties of the logistic sigmoid and its application to GRN ODEs; no new entities are postulated.

axioms (1)

standard math The logistic function is globally C^∞, real-valued for all real arguments, and strictly positive at zero
Invoked throughout the abstract as the basis for resolving the three Hill-function flaws.

pith-pipeline@v0.9.0 · 5622 in / 1343 out tokens · 45414 ms · 2026-05-09T14:34:04.551615+00:00 · methodology

discussion (0)

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Sustained Limit Cycles in the Logistic Two-Gene Genetic Oscillator: A Delay-Driven Hopf Bifurcation
math.DS 2026-05 unverdicted novelty 7.0

Delay-differential extension of the logistic two-gene oscillator undergoes a Hopf bifurcation at explicit critical total delay τ_c with closed-form ω_c, transversality bound, and supercritical amplitude laws, matching...
State-Feedback Control of Logistic-Based Gene Regulatory Networks: Closed-Form Lyapunov Certificates, Monostabilization, and Delay-Uniform Stability
math.DS 2026-06 unverdicted novelty 6.0

Introduces logistic-based GRN control with explicit feedforward-proportional laws, quadratic Lyapunov certificates for global exponential stability, monostabilization budgets, and delay-uniform stability theorems.
A Bias-Corrected Weighted Logistic Model for Gene Regulatory Networks: Functional Equivalence with the Product-of-Logistics and Comparison with Weighted-Sum Formulations
math.DS 2026-06 unverdicted novelty 6.0

Introduces bcw single-sigmoid model for GRNs that recovers product-of-logistics critical value 1/2^{m_i} and shared equilibrium, with Jacobian and stability comparisons.