Papex

Open problem · Biology · Open

Proteins of unknown function

What do the many conserved proteins whose biological role is still unknown actually do?

Summary

Two decades after the first eukaryotic genomes were sequenced, the biological role of around a fifth of their proteins is still unknown. Many of these proteins are conserved across species, which suggests they do something fundamental for life.

Why it is open

In 2019 Valerie Wood and colleagues found that about 20% of the proteins of human, budding yeast and fission yeast "lack physiologically informative descriptions", using a metric based on Gene Ontology terms for biological processes [8]. Many of these proteins are conserved from yeasts to human, "suggesting that they contribute to fundamental biological processes" [8]. In bacteria, Morgan Price and colleagues wrote in 2018 that one-third of all protein-coding genes "cannot be annotated with a function" [5].

"Unknown" has several meanings, and the numbers depend on which one is used [1, 3, 7]. A protein family can be unknown because no member has a known function; the Pfam database of protein families calls such families "domains of unknown function" (DUFs) [1]. A protein region can be "dark" in a structural sense, resembling no experimentally determined structure [3]. A protein can also carry a predicted function that no experiment has tested: Magdalena Antczak and colleagues noted in 2019 that "less than 1% of proteins have experimentally determined annotations" [7]. Predictions narrow the search, but David Bianchi and colleagues caution that structure "can offer hints as to the function of a protein" while further computational and experimental studies are required [11].

Which genes get studied is itself uneven: Thomas Stoeger and colleagues found that how much a human gene is studied can be explained to a large extent by a small set of its chemical, physical and biological properties [6].

No body sets a bar for a resolution. The problem shrinks one protein at a time as roles are tested; Wood and colleagues expect that closing the gap will need "a creative combination of existing and emerging experimental and in silico methods" [8].

What is known

Counting unknown families (2010–2021). Pfam labels a family a DUF when it contains no protein of known function and curators can find no functional information in the literature [1]. Alex Bateman, Penny Coggill and Robert Finn reported that by Pfam release 23.0 the numbering had reached DUF2607 and DUFs made up about 22% of all families; only 3.5% of DUF families were found in all kingdoms of life, against 15% of other families [1]. As of release 33.1, described in 2021, Pfam held 4,244 DUF or UPF (uncharacterised protein family) families, 23% of all families, and had renamed 1,132 such families because a function had been identified [9]. Norman Goodacre, Dietlind Gerloff and Peter Uetz found 238 DUFs in 355 essential proteins of 16 model bacteria, "clearly establishing the biological essentiality of DUFs" [2].

Conserved unknowns in model organisms (2019–2023). In fission yeast, Wood and colleagues identified 410 proteins of unknown physiological role with orthologues outside the Schizosaccharomyces clade; only 18 (4.4%) were essential under permissive growth conditions, against 29.2% of known proteins [8]. Among all 676 uncharacterised fission yeast genes, many more had growth or viability phenotypes under specific chemical challenges (26.1%) than under standard laboratory conditions (3.6%) [8]. In 2023 João Rocha and colleagues built the "Unknome database", which gives each protein a "knownness" score from its Gene Ontology annotations [14]. Among clusters containing human proteins, the share with a score of 2 or less fell from 43% to 23% over ten years, but knownness rose faster for genes that were already well annotated [14].

Minimal cells (2016–2022). In 2016 Clyde Hutchison and colleagues reported JCVI-syn3.0, a synthetic bacterium with 473 genes and a genome smaller than that of any autonomously replicating cell found in nature; unexpectedly, 149 of its genes had unknown biological functions [4]. By combining computational annotation methods, Antczak and colleagues assigned predicted functions to 66 of the 149, including 24 likely transporters [7]. For JCVI-syn3A, a version with 19 genes restored, Bianchi and colleagues reported in 2022 that about 90 of its 452 protein-coding genes (less than 20%) had no annotated function, in roughly equal thirds essential, quasi-essential and non-essential [11].

Large-scale screens (2018–2023). Price and colleagues measured the fitness of mutants of 32 bacteria across dozens of growth conditions and found phenotypes for 11,779 protein-coding genes that had not been annotated with a specific function [5]. Rocha and colleagues knocked down 260 unknown genes conserved between flies and humans in the fruit fly Drosophila by RNA interference; 62 (24%) compromised viability, and seven further assays found phenotypes for 59 more genes [14].

Predicted structures (2015–2023). Before large-scale structure prediction, Nelson Perdigão and colleagues defined the "dark proteome" as protein regions never observed by experimental structure determination and inaccessible to homology modelling; across 546,000 Swiss-Prot proteins, 44–54% of the proteome of eukaryotes and viruses was dark, against about 14% in archaea and bacteria [3]. By 2023 the AlphaFold database held over 214 million predicted structures [12]. Inigo Barrio-Hernandez and colleagues clustered them into 2.30 million non-singleton clusters, of which 31% lacked annotations; these covered only 4% of the proteins in the database [12]. Janani Durairaj and colleagues found that 34% of UniRef50 sequence clusters were "functionally dark", and 29% of those with a predicted structure [13].

Main approaches

Genetic screens. Disrupt every gene and find the conditions in which it matters [5]. Strongest evidence: the 11,779 bacterial genes with phenotypes, of which 2,316 had high-confidence associations conserved across bacteria, leading to proposed functions for enzymes, transporters and uncharacterised families [5]. Main difficulty: a phenotype is a lead, not a mechanism, and many conserved unknowns show phenotypes only under specific conditions [8]; RNA interference can have off-target effects [14].

Minimal cells. Study unknown genes in a cell with the fewest genes of unclear function [11]. Strongest evidence: using sequence analysis, structure prediction with the deep-learning method AlphaFold2 and interaction data, Bianchi and colleagues upgraded the annotations of 25 of the roughly 90 unclear JCVI-syn3A genes with high confidence and proposed targeted experiments [11]. Main difficulty: the proteins still unannotated lack orthologues or protein domains, or are membrane proteins [7].

Structure-guided discovery. Compare predicted structures to find distant relatives and new families [12, 13]. Strongest evidence: from a map of the AlphaFold database, Durairaj and colleagues described 290 putative new protein families and experimentally validated a new superfamily of toxin–antitoxin systems, TumE–TumA, whose toxin had been the family DUF6516 [13]; Barrio-Hernandez and colleagues found putative remote homology between human immune-related proteins and prokaryotic proteins [12]. Main difficulty: structure gives hints rather than function [11], and the unannotated structural clusters are small [12].

Machine-learning prediction. Train models on known families and predict functions for the rest [10]. Strongest evidence: in 2022 Maxwell Bileschi and colleagues at Google Research, with Bateman at EMBL-EBI, reported that deep learning models extend Pfam's coverage by more than 9.5%, "exceeding additions made over the last decade", and predict function for 360 human proteins with no previous Pfam annotation [10]; in 2025 Gemma Martínez-Redondo and colleagues reported that FANTASIA, which uses protein language models, predicts functions for virtually all proteins of about 1,000 animal proteomes, including up to 50% left unannotated by homology-based methods [15]. Main difficulty: both papers report predictions, and experimentally determined annotations cover less than 1% of proteins [7].

Open sub-questions

  1. Essential unknowns in JCVI-syn3A. For the essential genes of unclear function that remain after the 25 upgraded annotations [11], does depleting each one produce defects that match the role suggested by its predicted structure?
  2. Conditional phenotypes in fission yeast. Of the 410 conserved unknowns, few are essential under permissive conditions [8]; do the conditions that reveal phenotypes cluster around particular stresses that point to shared pathways?
  3. Testing deep-learning annotations. For the 360 human proteins given a first Pfam annotation by deep learning [10], how often do independent experimental data agree with the predicted family?
  4. Essential DUFs over time. Of the 238 DUFs found in essential bacterial proteins [2], how many have since been renamed in Pfam [9], and are essential DUFs characterised faster than others?
  5. Dark clusters and fitness data. Do high-confidence structural clusters without annotation [12] contain bacterial genes with strong mutant phenotypes [5], and do those phenotypes agree with structure-based predictions?
  6. Attention and knownness. Do the gene properties that predict research attention [6] also predict which conserved unknowns gained knownness after 2010 [14]?

Work on this in Papex

If you have a testable answer to one of the sub-questions above, you can draft it as a private hypothesis in Papex, or prepare the draft from Claude or ChatGPT with the Papex connector. You review the exact version, sign it and decide whether to publish it; a public record that names this problem is listed on this page. A Papex record shows who signed which version and when. It is not peer review and does not mean the hypothesis is correct.

Key references

  1. Bateman, A., Coggill, P. & Finn, R. D. (2010). DUFs: families in search of function. Acta Crystallographica Section F 66(10), 1148–1152. 10.1107/S1744309110001685 · Checked 16 Oct 2026
  2. Goodacre, N. F., Gerloff, D. L. & Uetz, P. (2014). Protein domains of unknown function are essential in bacteria. mBio 5(1), e00744-13 (published online 31 December 2013). 10.1128/mBio.00744-13 · Checked 16 Oct 2026
  3. Perdigão, N., Heinrich, J., Stolte, C., Sabir, K. S., Buckley, M. J., Tabor, B. et al. (2015). Unexpected features of the dark proteome. Proceedings of the National Academy of Sciences 112(52), 15898–15903. 10.1073/pnas.1508380112 · Checked 16 Oct 2026
  4. Hutchison, C. A., Chuang, R.-Y., Noskov, V. N., Assad-Garcia, N., Deerinck, T. J., Ellisman, M. H. et al. (2016). Design and synthesis of a minimal bacterial genome. Science 351(6280), aad6253. 10.1126/science.aad6253 · Checked 16 Oct 2026
  5. Price, M. N., Wetmore, K. M., Waters, R. J., Callaghan, M., Ray, J., Liu, H. et al. (2018). Mutant phenotypes for thousands of bacterial genes of unknown function. Nature 557(7706), 503–509. 10.1038/s41586-018-0124-0 · Checked 16 Oct 2026
  6. Stoeger, T., Gerlach, M., Morimoto, R. I. & Nunes Amaral, L. A. (2018). Large-scale investigation of the reasons why potentially important genes are ignored. PLOS Biology 16(9), e2006643. 10.1371/journal.pbio.2006643 · Checked 16 Oct 2026
  7. Antczak, M., Michaelis, M. & Wass, M. N. (2019). Environmental conditions shape the nature of a minimal bacterial genome. Nature Communications 10, 3100. 10.1038/s41467-019-10837-2 · Checked 16 Oct 2026
  8. Wood, V., Lock, A., Harris, M. A., Rutherford, K., Bähler, J. & Oliver, S. G. (2019). Hidden in plain sight: what remains to be discovered in the eukaryotic proteome? Open Biology 9(2), 180241. 10.1098/rsob.180241 · Checked 16 Oct 2026
  9. Mistry, J., Chuguransky, S., Williams, L., Qureshi, M., Salazar, G. A., Sonnhammer, E. L. L. et al. (2021). Pfam: the protein families database in 2021. Nucleic Acids Research 49(D1), D412–D419 (published online 2020). 10.1093/nar/gkaa913 · Checked 16 Oct 2026
  10. Bileschi, M. L., Belanger, D., Bryant, D. H., Sanderson, T., Carter, B., Sculley, D. et al. (2022). Using deep learning to annotate the protein universe. Nature Biotechnology 40(6), 932–937. 10.1038/s41587-021-01179-w · Checked 16 Oct 2026
  11. Bianchi, D. M., Pelletier, J. F., Hutchison, C. A., Glass, J. I. & Luthey-Schulten, Z. (2022). Toward the complete functional characterization of a minimal bacterial proteome. Journal of Physical Chemistry B 126(36), 6820–6834. 10.1021/acs.jpcb.2c04188 · Checked 16 Oct 2026
  12. Barrio-Hernandez, I., Yeo, J., Jänes, J., Mirdita, M., Gilchrist, C. L. M., Wein, T. et al. (2023). Clustering predicted structures at the scale of the known protein universe. Nature 622(7983), 637–645. 10.1038/s41586-023-06510-w · Checked 16 Oct 2026
  13. Durairaj, J., Waterhouse, A. M., Mets, T., Brodiazhenko, T., Abdullah, M., Studer, G. et al. (2023). Uncovering new families and folds in the natural protein universe. Nature 622(7983), 646–653. 10.1038/s41586-023-06622-3 · Checked 16 Oct 2026
  14. Rocha, J. J., Jayaram, S. A., Stevens, T. J., Muschalik, N., Shah, R. D., Emran, S. et al. (2023). Functional unknomics: systematic screening of conserved genes of unknown function. PLOS Biology 21(8), e3002222. 10.1371/journal.pbio.3002222 · Checked 16 Oct 2026
  15. Martínez-Redondo, G. I., Perez-Canales, F. M., Carbonetto, B., Fernández, J. M., Barrios-Núñez, I., Vázquez-Valls, M. et al. (2025). FANTASIA leverages language models to decode the functional dark proteome across the animal tree of life. Communications Biology 8, 1227. 10.1038/s42003-025-08651-2 · Checked 16 Oct 2026

Page reviewed 16 Oct 2026

Source

Open, as listed by Wikipedia · Unsolved problems in biology. Checked 08 Oct 2026.

The status is the source’s. Papex does not decide whether a problem is solved.

Records on this problem

No public record addresses this problem yet.

More in Life sciences