A recent framework (Torres Alegre 2026) treats proteomes as statistical languages and provides model-agnostic, cross-clade informational descriptors of protein sequence ensembles: composition entropy, adjacent mutual information, separation-dependent information profiles, and a compressibility measure, each compared against composition-matched null models. The descriptive layer is established; what the source explicitly leaves open is mechanistic grounding. Real proteomes depart from composition-matched i.i.d. and first-order Markov baselines, but the framework's author states that elevated long-range dependence does not uniquely identify a mechanism — heterogeneous protein-family mixture structure, repeats, and domain architecture can all generate it — and defers the disentangling analyses. The open problem is to quantitatively attribute the composition-independent informational structure of whole-proteome ensembles to its mechanistic sources.

Background

A proteome is the complete set of proteins encoded by an organism's genome; a reference proteome is a curated representative set of such sequences maintained in the UniProt database. Treating each protein simply as a string over the 20-letter amino-acid alphabet, a proteome becomes an ensemble of strings, and classical information theory — Shannon entropy, mutual information, and compressibility — quantifies its statistical structure without any assumption about folding or function.

In this vocabulary, the composition entropy H1=−∑xp1(x)log⁡2p1(x)H_1=-\sum_x p_1(x)\log_2 p_1(x) measures how unevenly amino acids are used, where p1(x)p_1(x) is the pooled one-site residue frequency; the adjacent mutual information I1I_1 measures how much knowing one residue reduces uncertainty about the next; and the information profile IdI_d extends this to residue pairs at separation dd along the chain, so its decay with dd characterizes the range of positional dependencies. Comparing a real ensemble to a ladder of null models — uniformly random strings; strings that are independent and identically distributed (i.i.d.) with the same composition; and first-order Markov chains (chains whose next-residue probabilities depend only on the current residue) matching the observed adjacent-pair statistics — separates the effects of mere composition from genuine positional structure. A gzip compression length per residue provides an orthogonal proxy for redundancy.

Torres Alegre (2026) built exactly this model-agnostic, cross-clade description: on 20 UniProt reference proteomes spanning 10 biological categories, with protein-level bootstrap resampling (resampling a proteome's proteins with replacement to quantify sampling variability) and matched synthetic controls, real proteomes were found to depart from composition-matched i.i.d. baselines (adjacent mutual information of roughly 0.0155 bits versus 0.0023 bits for controls), to exhibit information profiles that remain elevated above first-order Markov surrogates (synthetic sequences generated from the matched chains) at separations d>1d>1, and to be more compressible than i.i.d. controls by about 0.174 bits per character. The paper also proves an entropy-deficit bound: if only an exponentially small fraction of length-LL strings is viable, no viable ensemble can be simultaneously near-uniform in composition and near-i.i.d. in ordering. The work is explicit, however, that these signatures are descriptive fingerprints, not mechanistic explanations: the candidate sources it names for the positional structure are heterogeneous protein-family mixture structure (a proteome pools sequences from many distinct protein families, each with its own composition and local statistics), tandem repeats and low-complexity segments (segments of strongly biased or repetitive sequence), domain architecture (the arrangement of structural domains, the independent folding units of proteins, along a chain), and compositional segmentation (a sequence partitioned into blocks of distinct amino-acid usage).

The deferred mechanistic question has an unresolved prehistory. Weiss, Jiménez-Montaño and Herzel (2000) measured entropy and compression of non-redundant protein sequence sets against random surrogates, found only about 1% entropy reduction from correlations, and could only speculate that secondary structure and low-complexity regions cause it. Cai, Dodson, Doig and Wolkenhauer (2002) detected amino-acid self-clustering. Grzymski and Marsh (2014) showed that amino-acid usage differs with microbial lifestyle — a composition-level effect. Model-based lines such as Direct Coupling Analysis, which infers pairwise residue couplings within protein families using Potts models (maximum-entropy models over sequence families; Weigt et al. 2009), and trained protein language models capture positional statistics, but only within families or inside a fitted model, not as model-agnostic, whole-proteome attribution.

Problem Statement

The descriptive framework exists; the open problem is its mechanistic grounding. For whole-proteome sequence ensembles, determine the origin of the composition-independent informational structure observed across clades — the excess of adjacent mutual information over composition-matched i.i.d. baselines, the elevation of the information profile IdI_d above first-order Markov surrogates at separations d>1d>1, and the compressibility gap relative to i.i.d. controls at matched composition — by quantitatively decomposing these signatures into their candidate sources: heterogeneous protein-family mixture structure, tandem repeats and low-complexity segments, domain architecture, compositional segmentation, and any residual within-protein ordering constraints. The source itself defers the stratified and controlled analyses that could support such a decomposition and estimator-bias and subsampling-sensitivity treatment adequate for cross-clade comparison; the single objective uniting these components is a determination of whether the signatures are accounted for by heterogeneous protein-family mixture structure, repeats and low-complexity segments, domain architecture, and compositional segmentation, or whether a genuine global (within-protein) ordering-constraint component remains. The question is posed at the whole-proteome, cross-clade scope of the source, without restriction to particular taxa, separations, or answer forms.

References

  1. Enso O. Torres Alegre (2026), 'Proteins as Statistical Languages: Information-Theoretic Signatures of Proteomes Across the Tree of Life', bioRxiv preprint posted 2026-01-05, DOI 10.64898/2025.12.31.697235, https://doi.org/10.64898/2025.12.31.697235
  2. Olaf Weiss, Miguel A. Jiménez-Montaño, Hanspeter Herzel (2000), 'Information content of protein sequences', Journal of Theoretical Biology 206(3):379–386, DOI 10.1006/jtbi.2000.2138, https://doi.org/10.1006/jtbi.2000.2138
  3. Yudong Cai, C. T. J. Dodson, Andrew J. Doig, Olaf Wolkenhauer (2002), 'Information-Theoretic Analysis of Protein Sequences Shows that Amino Acids Self-cluster', Journal of Theoretical Biology 218(4):409–418, DOI 10.1006/jtbi.2002.3088, https://doi.org/10.1006/jtbi.2002.3088
  4. Joseph J. Grzymski, Adam G. Marsh (2014), 'Protein Languages Differ Depending on Microorganism Lifestyle', PLoS ONE 9(5):e96910, DOI 10.1371/journal.pone.0096910, https://doi.org/10.1371/journal.pone.0096910
  5. Martin Weigt, Robert A. White, Hendrik Szurmant, James A. Hoch, Terence Hwa (2009), 'Identification of direct residue contacts in protein–protein interaction by message passing', Proceedings of the National Academy of Sciences USA 106(1):67–72, DOI 10.1073/pnas.0805923106, https://doi.org/10.1073/pnas.0805923106
  6. Anirjit Chatterjee, Nagarjun Vijay (2026), 'A benchmarking framework for comparative evaluation of low-complexity region detection tools in the human proteome', Scientific Reports 16:17498, DOI 10.1038/s41598-026-48432-3, https://doi.org/10.1038/s41598-026-48432-3
  7. The UniProt Consortium (Alex Bateman, Maria-Jesus Martin, Sandra Orchard, Michele Magrane, Aduragbemi Adesina, Shadab Ahmad, Emily H. Bowler-Barnett, Hema Bye-A-Jee, David Carpentier, Paul Denny, Jun Fan, Penelope Garmiri, Leonardo Jose da Costa Gonzales, Abdulrahman Hussein, Alexandr Ignatchenko, Giuseppe Insana, Rizwan Ishtiaq, Vishal Joshi, Dushyanth Jyothi, Swaathi Kandasaamy, Antonia Lock, Aurelien Luciani, Jie Luo, Yvonne Lussi, Juan Sebastian Martinez Marin, Pedro Raposo, Daniel L. Rice, Rafael Santos, Elena Speretta, James Stephenson, Prabhat Totoo, Nidhi Tyagi, Nadya Urakova, Preethi Vasudev, Kate Warner, Supun Wijerathne, Conny Wing-Heng Yu, Rossana Zaru, Alan J. Bridge, Lucila Aimo, Ghislaine Argoud-Puy, Andrea H. Auchincloss, Kristian B. Axelsen, Parit Bansal, Delphine Baratin, Teresa M. Batista Neto, Marie-Claude Blatter, Jerven T. Bolleman, Emmanuel Boutet, Lionel Breuza, Blanca Cabrera Gil, Cristina Casals-Casas, Kamal Chikh Echioukh, Elisabeth Coudert, Beatrice Cuche, Edouard de Castro, Anne Estreicher, Maria L. Famiglietti, Marc Feuermann, Elisabeth Gasteiger, Pascale Gaudet, Sebastien Gehant, Vivienne Gerritsen, Arnaud Gos, Nadine Gruaz, Chantal Hulo, Nevila Hyka-Nouspikel, Florence Jungo, Arnaud Kerhornou, Philippe Le Mercier, Damien Lieberherr, Patrick Masson, Anne Morgat, Salvo Paesano, Ivo Pedruzzi, Sandrine Pilbout, Lucille Pourcel, Sylvain Poux, Monica Pozzato, Manuela Pruess, Nicole Redaschi, Catherine Rivoire, Christian J. A. Sigrist, Karin Sonesson, Shyamala Sundaram, Anastasia Sveshnikova, Cathy H. Wu, Cecilia N. Arighi, Chuming Chen, Yongxing Chen, Hongzhan Huang, Kati Laiho, Minna Lehvaslaiho, Peter McGarvey, Darren A. Natale, Karen Ross, C. R. Vinayaka, Yuqi Wang, Jian Zhang) (2025), 'UniProt: the Universal Protein Knowledgebase in 2025', Nucleic Acids Research 53(D1):D609–D617, DOI 10.1093/nar/gkae1010, https://doi.org/10.1093/nar/gkae1010