Dark proteome
The proteins science has read but barely understands. Many come from microbes in places most of us will never see.
The dark proteome is the part of the protein world we know almost nothing about. The term first meant proteins with no known shape. Since AlphaFold, people use it more for proteins with no known function.
Two meanings
The term took hold in 2015, when Nelson Perdigão and colleagues surveyed the “dark” parts of proteins: stretches never seen in an experiment, with no known relative to model them from. Across about 546,000 well-studied proteins, they found that between 44 and 54 percent of the proteome in animals, plants, fungi and viruses was dark. In bacteria and archaea it was about 14 percent.
Then AlphaFold predicted shapes for more than 200 million proteins. Much of the darkness in shape lifted, at least as prediction. What stayed dark was function. In 2023 one study found that about a third of all groups of related proteins in UniProt were “functionally dark”, with almost nothing known about what they do. Another grouped AlphaFold’s predicted shapes into 2.3 million families and found that 31% matched no known structure or protein family.
So the word is now used both ways. When someone says “dark proteome”, it’s worth asking which one they mean.
Where dark proteins come from
Many come from microbes that have never been grown in a lab. Scientists know them only from DNA read straight out of soil, seawater, hot springs and the sea floor. A 2023 study searched DNA from almost 27,000 such samples and found 1.17 billion protein sequences unlike anything in the reference collections. Grouped into families, they doubled the number of protein families known from reference genomes.
Dark doesn’t mean unimportant. Digging into dark families, the 2023 AlphaFold study found a toxin and antitoxin pair in bacteria nobody had described before, and confirmed it in the lab.
Why designers should care
Dark proteins are tempting. They’re new, unnamed and unclaimed, and in the AlphaFold Database many of them look finished: a confident shape, in colour, with nothing to say what it does.
A shape is a guess at structure. It says little about what a protein does in a living cell, what it needs, or what it touches. Reading that gap is the judgement part of being an AI-native biodesigner.
There is a care question too. Dark proteins often come from samples taken in specific places. The questions about digital sequence information apply here, and they are hardest to answer where we know least.
What you can do as a designer
Read the name and the evidence. “Uncharacterized protein” or “domain of unknown function” in a UniProt entry means what it says.
Treat any function you read into a shape as a guess. Write it down as a question to test.
Check the prediction’s confidence. AlphaFold colours each part of the chain by how sure it is.
Find where the sample came from. Note the organism or environment, and the place if it’s recorded. See data provenance.
Say “function unknown” out loud in your work. It is a finding, and an honest place to start.
Perdigão, N. et al. (2015). Unexpected features of the dark proteome. PNAS, 112(52), 15898–15903. doi.org/10.1073/pnas.1508380112
Durairaj, J. et al. (2023). Uncovering new families and folds in the natural protein universe. Nature, 622, 646–653. doi.org/10.1038/s41586-023-06622-3
Barrio-Hernandez, I. et al. (2023). Clustering predicted structures at the scale of the known protein universe. Nature, 622, 637–645. doi.org/10.1038/s41586-023-06510-w
Pavlopoulos, G. A. et al. (2023). Unraveling the functional dark matter through global metagenomics. Nature, 622, 594–602. doi.org/10.1038/s41586-023-06583-7
