Modeling Discrete Combinatorial Systems as Alphabetic Bipartite Networks: Theory and Applications

Life and language are discrete combinatorial systems (DCSs) in which the basic building blocks are finite sets of elementary units: nucleotides or codons in a DNA sequence and letters or words in a language. Different combinations of these finite units give rise to potentially infinite numbers of genes or sentences. This type of DCS can be represented as an Alphabetic Bipartite Network ($\alpha$-BiN) where there are two kinds of nodes, one type represents the elementary units while the other type represents their combinations. There is an edge between a node corresponding to an elementary unit $u$ and a node corresponding to a particular combination $v$ if $u$ is present in $v$. Naturally, the partition consisting of the nodes representing elementary units is fixed, while the other partition is allowed to grow unboundedly. Here, we extend recently analytical findings for $\alpha$-BiNs derived in [Peruani et al., Europhys. Lett. 79, 28001 (2007)] and empirically investigate two real world systems: the codon-gene network and the phoneme-language network. The evolution equations for $\alpha$-BiNs under different growth rules are derived, and the corresponding degree distributions computed. It is shown that asymptotically the degree distribution of $\alpha$-BiNs can be described as a family of beta distributions. The one-mode projections of the theoretical as well as the real world $\alpha$-BiNs are also studied. We propose a comparison of the real world degree distributions and our theoretical predictions as a means for inferring the mechanisms underlying the growth of real world systems.

💡 Research Summary

This paper introduces Alphabetic Bipartite Networks (α‑BiNs) as a formal framework for representing discrete combinatorial systems (DCSs) such as genetic sequences and natural languages. In an α‑BiN there are two partitions of nodes: a fixed set U of elementary units (e.g., nucleotides, codons, letters, phonemes) and a growing set V of their discrete combinations (genes, sentences, languages). An edge connects a unit u∈U to a combination v∈V whenever u appears in v; multiple occurrences are modeled as multi‑edges or weighted edges. The authors extend earlier analytical work on α‑BiNs and derive growth equations for two attachment regimes: sequential attachment (μ = 1, one edge per time step) and parallel attachment (μ > 1, μ edges added simultaneously). Parallel attachment is further split into “with replacement” (allowing multi‑edges) and “without replacement” (single edges only).

The attachment probability follows a kernel that mixes preferential attachment with a tunable randomness parameter γ (or equivalently an initial attractiveness α = 1/γ):
e_A(k) = (γk + 1)/(γμt + N), where N = |U| and k is the current degree of a unit node. This kernel reduces to pure preferential attachment when γ→∞ and to uniform random attachment when γ→0.

Using a master‑equation approach, the degree distribution p_{k,t} of the fixed partition U is approximated by
p_{k,t+1} = (1 − A_p(k,t)) p_{k,t} + A_p(k−1,t) p_{k−1,t},
with A_p(k,t) = (γk + 1)μ/(γμt + N). Solving this recurrence yields a closed‑form expression involving products of gamma‑like terms, which asymptotically converges to a family of beta distributions. Depending on γ, four characteristic shapes emerge: (a) γ = 0 → binomial/Poisson‑like, (b) 0 < γ < 1 → skewed normal with a moving peak, (c) 1 ≤ γ ≤ (N/μ) − 1 → monotonically decreasing (exponential‑like), and (d) γ > (N/μ) − 1 → U‑shaped with high probability at both extremes. Simulations confirm these theoretical predictions for a range of parameters.

To validate the theory, the authors construct two empirical α‑BiNs. The first is a codon‑gene network where U consists of the 64 possible codons and V comprises genes from several organisms (bacteria, yeast, human). Each gene is a multiset of codons, so the “with replacement” parallel attachment model applies. Fitting the model to the observed degree distributions yields γ values that increase with organismal complexity, indicating a higher degree of randomness in codon usage for more complex genomes.

The second empirical system is a phoneme‑language network: U contains all phonemes observed across a sample of world languages, and V represents individual languages as sets of phonemes. Here the “without replacement” parallel attachment model is appropriate because a language cannot contain the same phoneme multiple times. The observed phoneme degree distribution deviates from the predicted U‑shaped beta distribution, showing instead a heavy tail for a few highly frequent phonemes. Moreover, the one‑mode projection onto the phoneme nodes (i.e., a phoneme‑phoneme co‑occurrence network) exhibits topological features not captured by the simple preferential‑attachment model. This discrepancy suggests that additional mechanisms—such as phonotactic constraints, cultural diffusion, and functional load—play a significant role in shaping language phoneme inventories.

The paper also examines the one‑mode projection of α‑BiNs onto the fixed partition (U‑U networks). While the theoretical degree distribution of the projected network can be derived from the beta‑distributed original degrees, empirical projections (especially for the phoneme‑language case) display systematic differences, highlighting the limitations of the current growth model.

In the discussion, the authors acknowledge several simplifications: (1) the order of basic units within a combination is ignored (the model treats words as bags of letters and genes as multisets of codons), (2) the randomness parameter γ is assumed constant over time, and (3) only two partitions are considered, whereas many real systems involve multiple interacting layers (e.g., genes, proteins, metabolic pathways). They propose extensions such as incorporating ordered sequences, allowing γ to evolve, and building multilayer α‑BiNs to capture richer dynamics.

Overall, the study provides a mathematically tractable framework for analyzing DCSs, demonstrates that the degree distribution of the elementary‑unit side follows a beta‑distribution family, and shows that fitting empirical data can reveal underlying growth mechanisms. However, the mismatch observed in the phoneme‑language projection underscores the need for more sophisticated models that account for domain‑specific constraints and interactions.