HN Debrief

The Unreasonable Redundancy of Nature's Protein Folds

  • Biology
  • AI
  • Machine Learning
  • Research
  • Synthetic Biology

The post says nature appears to overuse a limited catalog of protein folds, meaning many very different protein sequences and functions collapse onto the same small family of 3D structures. In plain terms, proteins are chains of amino acids that fold into shapes, and those shapes often matter more for function than exact sequence. The claim is not that proteins are all similar, but that biology keeps reaching for the same structural solutions. That matters because it bears on a live question in protein science and AI biology: is fold space genuinely small because physics allows only a limited number of stable, useful architectures, or is natural biology stuck in a historically accessible subset because evolution mostly modifies what already exists.

If you work on protein design or AI for biology, do not treat the set of known natural folds as the full space of what can work. The practical opportunity is in tools and experiments that push beyond evolution's reused templates, especially with designed proteins, altered cellular machinery, or non-canonical amino acids.

Discussion mood

Interested but not dazzled. The dominant reaction was that the core observation matches long-standing biochemistry knowledge, while the useful part is the sharper question it raises about whether nature's small fold catalog reflects physical limits, evolutionary path dependence, or both.

Key insights

  1. 01

    Accessible fold space is not total fold space

    What matters here is the gap between what can exist and what evolution can realistically reach. Sequence space is astronomically large, but folds only matter if there is a broad enough basin in sequence space for evolution to stumble into and keep. That reframes the post from a cataloging claim into a search problem. Nature's redundancy may say more about accessibility and path dependence than about a hard ceiling on possible folds.

    Treat natural proteins as samples from an accessible region, not as an exhaustive library. If you are designing proteins, bias your search toward robustness and reachable mutational neighborhoods, then test whether engineered systems can escape them.

      Attribution:
    • DrScientist #1 #2
    • Windchaser #1
    • pfdietz #1
  2. 02

    AlphaFold cannot answer the whole possibility question

    AlphaFold-style models are bounded by the world they were trained on. They learn from natural protein sequences and known structures, so they are strong at recovering biology's existing patterns but weak as tools for asking what folds are possible with different amino-acid alphabets, modified proteins, or folding environments that nature rarely uses. That makes them excellent interpolation engines, not final arbiters of protein possibility space.

    Do not use high confidence from a structure predictor as evidence that unexplored designs are impossible. For novel chemistries or far-from-natural scaffolds, plan for direct experiments or design methods that are not anchored to the natural training set.

      Attribution:
    • photochemsyn #1
    • flobosg #1
  3. 03

    Cellular machinery may gate which folds are reachable

    Fold space may be limited not just by thermodynamics but by what a cell can afford to build and fold. Chaperones, cofactors, ribosomal exit dynamics, and other cellular machinery can make some structures practical and leave others effectively unreachable. That pushes the question beyond pure protein physics. A different cellular toolkit might unlock structures that natural evolution almost never accesses.

    If you want truly new protein architectures, change the environment as well as the sequence. Synthetic biology platforms with altered chaperones, cofactors, or compartment-like helpers may be as important as better sequence design.

      Attribution:
    • jeejay1 #1
    • gilleain #1 #2
    • hirenj #1
  4. 04

    Reuse happens below the whole-fold level too

    The redundancy is not only at the level of complete folds. Protein scientists in the comments pointed out that small structural fragments and motifs recur across folds that are otherwise treated as evolutionarily distant. Rossmann folds and TIM barrels are the famous examples at the fold level, but the more interesting point is that biology seems to reuse sub-domain building blocks over and over as well.

    Think modularly when building protein tools. Useful novelty may come from recombining proven fragments and active-site motifs, not only from inventing entirely new top-level folds.

      Attribution:
    • flobosg #1 #2
    • resiros #1
  5. 05

    Function does not require a rigid final structure

    The fold conversation can mislead if it assumes useful proteins must settle into one clean 3D shape. Intrinsically disordered proteins and proteins from de novo genes often work while remaining disordered or only partly structured, sometimes in molten-globule-like states. That means a bounded catalog of stable folds does not bound the full space of functional proteins.

    Do not scope protein discovery too narrowly around crisp folded domains. If your application involves signaling, regulation, or transient interactions, include disordered and partially structured candidates in the search.

      Attribution:
    • dekhn #1
    • flobosg #1

Against the grain

  1. 01

    The result is old news to biochemists

    The strongest pushback was that recurring protein motifs and strong structure conservation despite sequence drift have been standard teaching for decades. From that view, the methods may be new but the headline overstates the surprise. The interesting unanswered part is not whether biology reuses folds, but how much unseen fold space still exists beyond the recurring motifs everyone already knew about.

    Read this as a better quantified restatement of a familiar pattern, not as a discovery that overturns protein science. If you communicate similar work, emphasize what is newly measured or newly enabled instead of selling recurrence itself as a shock.

  2. 02

    Older structure catalogs were experimentally biased

    A useful correction to the “we knew this 30 years ago” argument is that old structure databases were shaped by what crystallized and could be measured easily. Recurrent folds in those datasets were real, but the apparent size of fold space was also filtered by experimental tractability. That keeps the door open to the idea that biology contains more structural variety than the classic literature suggested.

    Be careful when using historical structural databases as evidence that protein architecture is mostly solved. Check how much your conclusion depends on measurement bias versus true biological prevalence.

      Attribution:
    • DrScientist #1

In plain english

AlphaFold ↩
An artificial intelligence system that predicts a protein's three-dimensional structure from its amino-acid sequence.
amino acids ↩
The small molecular building blocks that link together to make proteins.
chaperones ↩
Cellular helper proteins that assist other proteins in folding correctly.
cofactors ↩
Non-protein molecules or ions that some proteins need in order to function properly.
de novo genes ↩
Genes that appear to have arisen newly rather than by modifying an older existing gene.
intrinsically disordered proteins ↩
Proteins or protein regions that do not settle into one stable 3D structure yet can still perform biological functions.
non-canonical amino acids ↩
Amino acids outside the standard set normally used by living cells to build proteins.
sequence space ↩
The enormous set of all possible amino-acid sequences a protein could have.
thermodynamics ↩
The physical rules governing energy and stability that help determine whether a protein shape can form and persist.

Reference links

Protein fold and structure references

  • Rossmann fold
    Given as a classic example of one structural template reused for many functions.
  • De novo protein folds review
    Used to ground the claim that even a small corner of fold space may contain on the order of a thousand possible folds.
  • Protein length reference
    Cited during a side discussion about whether longer proteins would meaningfully expand the number of folds.
  • Vault organelle
    Mentioned as an example of unusual cellular machinery that could in principle help access different folding possibilities.

Evolution and protein design framing

Beyond canonical proteins

Related talks and essay references