bioRxiv Science⌕ Search

bioRxiv · 10.1101/2025.08.23.671922

Individual-based modeling of direct and indirect competition by the glass-box artificial intelligence

Abstract

The main drawback of traditional ecological models is the inability to model local interactions between individuals. As a consequence, it was impossible to mathematically model individual-based mechanisms of direct and indirect competition, which hampers the understanding of ecological dynamics. To overcome this problem, we developed a fully transparent mechanistic analog of the classical Lotka-Volterra phenomenological model based on explicitly explainable artificial intelligence (AI) of the glass-box type. In this paper, on an individual-based model, we present (i) a rigorous mechanism of inverted competitive exclusion where, all else being equal, the weaker species in direct competition defeats the stronger species through indirect exploitative competition and (ii) a rigorous mechanism of coexistence of complete competitors resulting from a combination of accounting for microhabitat regeneration, moderate reproduction, a certain habitat size and boundary conditions. Both presented mechanisms go beyond the classical formulations of the competitive exclusion principle, as in our model two species with different competitiveness compete for a single limiting resource in one homogeneous bounded ecosystem. We have shown that individual-based modeling of direct and indirect competition mechanisms allows for a more detailed understanding of the causal mechanisms of interspecific competition. Such understanding is necessary to refine formulations of the competitive exclusion principle, to search for mechanisms of competitive coexistence, to predict ecosystem behavior, and to develop conservation strategies. Our approach allows us to study discrete systems with fundamentally discrete mathematics, which opens up new opportunities for achieving accuracy, clarity, rigor, and explainability in studies of complex systems. HighlightsO_LIMechanisms of direct and indirect competition in one individual-based model C_LIO_LIA mechanism of indirect exploitative competition by which a weak competitor wins C_LIO_LIA mechanism of coexistence of complete competitors C_LIO_LICellular automata for the eXplicitly eXplainable AI (XXAI) of the glass-box type C_LIO_LIODD protocol of the model and a source code of the Monte Carlo experiment in C++ C_LI

Source connections

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Kalmykov, V. L., Kalmykov, L. V.. 2025-08-28. Individual-based modeling of direct and indirect competition by the glass-box artificial intelligence. https://doi.org/10.1101/2025.08.23.671922

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

The impacts of exogenous noise on stochastic disease dynamics

Much of the literature and intuition associated with mathematical epidemiology is driven by deterministic models, which are a reasonable assumption when the population size is large. Stochastic models, especially individual based models, are however considered vital when dealing with small population sizes, especially at times of invasion or extinction. The overwhelming majority of these models (both deterministic and stochastic) assume that the underlying parameters are fixed (or follow a regular seasonal pattern). Here, we consider an analytic framework for dealing with randomly varying parameters through the use of stochastic differential equations - thereby capturing the action of external noisy processes such as weather. In particular, we focus on when the transmission rate, {beta}, varies as the solution to a Cox-Ingersoll-Ross Model, such that {beta} is gamma distributed with autocorrelation. We consider the impact of this parameter variation on a stochastic version of the Susceptible-Infected-Recovered model, and for this 'double-stochastic' model show through simulation and analytical results that exogenous noise increases the impact of stochasticity, potentially leading to more early extinctions, wider variations in the number of cases at equilibrium, but that early growth rate can be faster or slower depending on the precise parameters.

systems biology↗

Site-resolved spatial and structural interactome of a human cell

The spatial and structural arrangement of proteins determine virtually every process in human cells. We combined gentle subcellular fractionation by differential ultracentrifugation with cross-linking mass spectrometry to systematically map this cellular proteome architecture with residue-level evidence, identifying 164,146 residue-to-residue links in HEK293 cells. These links capture spatial protein arrangement at a resolution sufficient to pinpoint protein localizations at sub-organelle level, determine protein orientations within cellular membranes, and identify inter-organelle contact sites. The residue-level information provides evidence for 18,074 direct protein-protein interactions (PPIs), which we integrate into AlphaFold-based pipelines to nominate PPI-mediating short linear motifs and generate assembly models of large protein complexes. Guided by these spatial and structural readouts, we discover new PPIs within the endomembrane system that regulate the compartmental localization of trafficking machinery. Leveraging a network topology-driven strategy, we augment our HEK293 dataset with PPI data from different cell lines and methods, expanding the spatial and structural interactome of human cells.

systems biology↗

Sensory variation and behavioural degeneracy: a framework for interpreting heterogeneity in the gut-brain axis

Gut microbiome differences are frequently interpreted as reflecting underlying biological differences between individuals. When outcomes are mediated by behaviour, however, this mapping may be fundamentally non-unique. This limits causal inference in gut-brain research, where microbiome differences in autism and depression are routinely attributed to intrinsic neurobiology despite highly variable, overlapping findings. I built a minimal agent-based model grounded in the known sensory variation across the autism spectrum. Dietary behaviour emerges from latent sensory traits, including sensory drive, predictability preference, and context sensitivity, through reinforcement learning and environmental interaction. This behaviour shapes gut microbiome composition. Behavioural variation organizes endogenously into a continuum of specialist, opportunist, and explorer strategies that maps onto the autism sensory spectrum. The system is fundamentally degenerate. Similar microbiome states arise from distinct behavioural pathways. Similar dietary patterns emerge from divergent latent traits. This many-to-one mapping reflects the structural interaction of behaviour, learning, and environmental variability, not stochasticity alone. Microbiome similarity therefore does not uniquely identify underlying cause. As such, the model provides a theoretical framework for interpreting heterogeneity in gut microbiome research, particularly in autism, and generalizes to any condition where behaviour mediates between neural processes and ecological outcomes.

systems biology↗