The proteins that do not hold still

Introduction

AlphaFold is often presented as a landmark example of AI accelerating science. It can predict the three-dimensional structures of many proteins with striking accuracy, helping researchers explore biology far faster than before. Yet some of the most medically important proteins expose a fundamental limit in that success story: they do not settle into a single stable shape at all.

These molecules are known as intrinsically disordered proteins, or proteins with intrinsically disordered regions. Rather than behaving like rigid molecular machines, they shift between many conformations, changing shape depending on their environment and interaction partners. That flexibility is not a minor exception. Disordered regions are common in human biology and are heavily involved in gene regulation, signalling, ageing, neurodegeneration and cancer.[PMC]pmc.ncbi.nlm.nih.govComparative evaluation of AlphaFold2 and disorder predictors…by B Zhao · 2023 · Cited by 45 — Intrinsically disordered proteins (ID…

For AlphaFold, they function as a stress test. The system was built around predicting structure, but some biologically important proteins are defined precisely by the absence of one fixed structure. Understanding where AlphaFold struggles helps clarify both the promise and the limits of AI-driven scientific acceleration. It shows that some of biology’s hardest problems are not merely questions of computing power, but questions about how living systems remain dynamic, context-dependent and constantly in motion.

The proteins that do not hold still

Traditional structural biology often treats proteins as objects that fold into a specific three-dimensional arrangement. That picture works well for many enzymes and structural proteins. Their function depends on maintaining a relatively stable shape.

Intrinsically disordered proteins (IDPs) and intrinsically disordered regions (IDRs) are different. Under normal physiological conditions, they do not adopt a single long-lived structure. Instead, they exist as shifting ensembles of possible conformations. The protein may briefly sample one arrangement before moving to another, with no single configuration dominating.[PMC]pmc.ncbi.nlm.nih.govSystematic identification of conditionally folded intrinsically…by TR Alderson · 2023 · Cited by 246 — We found that AlphaFold2 can…

This apparent disorder is not necessarily a defect. Evolution has repeatedly preserved these flexible regions because they can perform tasks that rigid proteins struggle to accomplish. Their flexibility allows them to:

Large fractions of the eukaryotic proteome contain substantial disordered regions. Estimates commonly suggest that roughly one-third of eukaryotic proteins contain long intrinsically disordered segments.[PMC]pmc.ncbi.nlm.nih.govAlphaFold2: A Role for Disordered Protein/Region Prediction?by CJ Wilson · 2022 · Cited by 183 — We assess the ability of AlphaFold2 t…

For a system like AlphaFold, that creates an unusual challenge. Predicting a single stable structure is no longer the right objective if biological reality is a constantly changing ensemble.

Why disorder matters for ageing and disease

The importance of disordered proteins becomes clearer when looking at diseases associated with ageing.

Many neurodegenerative disorders involve proteins that are either intrinsically disordered or contain large disordered regions. Proteins associated with Alzheimer’s disease, Parkinson’s disease, Huntington’s disease and related disorders can transition from flexible states into pathological aggregates. The disease process often depends less on one stable structure than on a complicated landscape of transient intermediates, misfolded assemblies and self-organising aggregates.[ScienceDirect]sciencedirect.comAlphaFold and the amyloid landscapeby F Pinheiro · 2021 · Cited by 95 — In this review, we discuss the applications and limi…

This is one reason why protein prediction alone does not solve neurodegeneration. A static model may capture a plausible configuration, yet the disease may emerge from rare conformations, aggregation pathways or dynamic interactions that occur over time.

Disordered proteins also appear throughout cancer biology. Many transcription factors, signalling proteins and regulatory hubs contain substantial disordered regions. These flexible segments help coordinate cellular decisions, but they can also become points of failure when mutations alter signalling networks.[PMC]pmc.ncbi.nlm.nih.govAlphaFold‐pLDDT uses 1 ‐ pLDDT as output.Read more…

Ageing itself increasingly appears connected to failures in protein quality control, aggregation management and cellular regulation. Since intrinsically disordered proteins are deeply involved in these processes, understanding them could be important for any future effort to extend healthy lifespan or reduce age-related disease burden. Yet they are precisely the class of proteins least amenable to simple structure prediction.[arXiv]arxiv.orgSource details in endnotes.

This creates a useful reality check for broader claims about AI-enabled longevity. Accelerating structural biology is valuable, but some of the proteins most relevant to ageing are difficult because they are not static objects waiting to be mapped. They are dynamic systems whose behaviour depends on motion, context and interaction.

Why AlphaFold struggles with disorder

AlphaFold’s achievement was extraordinary because it learned statistical relationships between amino-acid sequences and experimentally determined structures. In many cases, it can predict a protein’s folded state with accuracy approaching laboratory methods.[Nature]nature.comAtomic resolution ensembles of intrinsically disordered…by V Schnapka · 2026 · Cited by 17 — Intrinsically disordered proteins are ubi…

The problem is that intrinsically disordered proteins do not necessarily possess one dominant folded state.

Researchers studying AlphaFold and disorder have repeatedly noted that the system’s strongest predictions occur when a protein resembles the kinds of folded structures that dominate structural databases. When proteins remain highly dynamic, the prediction problem changes fundamentally. ScienceDirect[EMBL-EBI]pmc.ncbi.nlm.nih.govAlphaFold‐pLDDT uses 1 ‐ pLDDT as output.Read more…

A useful analogy is the difference between predicting the shape of a building and predicting the movements of a crowd. A building has a stable geometry. A crowd constantly rearranges itself. The challenge is no longer identifying a single correct configuration but characterising a distribution of possibilities.

This limitation appears throughout the AlphaFold literature. Reviews of the system consistently identify intrinsically disordered regions as one of the major remaining weaknesses of current structure-prediction approaches. PMC[Springer Link]link.springer.comSpringer LinkAdvantages and Limitations of AlphaFold in Structural Biologyby MQC Li · 2025 · Cited by 2 — A key limitation is AlphaFold's…

Even AlphaFold’s own educational materials emphasise that the system cannot truly predict dynamic substructures that lack a fixed natural conformation.[EMBL-EBI]pmc.ncbi.nlm.nih.govAlphaFold‐pLDDT uses 1 ‐ pLDDT as output.Read more…

What confidence scores can and cannot tell us

One reason AlphaFold remains useful despite these limitations is that it usually signals when it is uncertain.

The system produces a confidence measure known as pLDDT, or predicted Local Distance Difference Test. Higher scores generally indicate greater confidence that the predicted local structure is accurate. Lower scores indicate uncertainty. Nature[EMBL-EBI]pmc.ncbi.nlm.nih.govAlphaFold‐pLDDT uses 1 ‐ pLDDT as output.Read more…

In practice, many intrinsically disordered regions receive low confidence scores. Researchers quickly realised that this behaviour could itself be informative. Instead of viewing low-confidence regions purely as failures, scientists began using them as clues that a protein might contain disorder.[PMC]pmc.ncbi.nlm.nih.govComparative evaluation of AlphaFold2 and disorder predictors…by B Zhao · 2023 · Cited by 45 — Intrinsically disordered proteins (ID…[PMC]pmc.ncbi.nlm.nih.govSystematic identification of conditionally folded intrinsically…by TR Alderson · 2023 · Cited by 246 — We found that AlphaFold2 can…

However, an important distinction is often lost in public discussions.

Low confidence does not automatically mean disorder.

A low pLDDT score can reflect several possibilities:

The reverse problem also exists. Some disordered regions can occasionally receive predictions that appear more structured than biological experiments would suggest. Researchers have found cases where AlphaFold identifies conditionally folded regions that become structured only in specific contexts or interactions.[PMC]pmc.ncbi.nlm.nih.govAdvantages and Limitations of AlphaFold in Structural Biologyby MQC Li · 2025 · Cited by 7 — However, AlphaFold also exhibits intrinsi…

This means confidence scores are useful indicators rather than definitive answers. They help researchers identify where caution is required, but they do not replace experimental evidence.

When AlphaFold succeeds anyway

The story is not simply one of failure.

Researchers have discovered that AlphaFold can sometimes reveal important information about disorder precisely because it was not designed as a disorder predictor. Studies comparing AlphaFold outputs with specialised disorder-prediction methods found that the system can often discriminate between ordered and disordered regions surprisingly well.[PMC]pmc.ncbi.nlm.nih.govAlphaFold2: A Role for Disordered Protein/Region Prediction?by CJ Wilson · 2022 · Cited by 183 — We assess the ability of AlphaFold2 t…

Some apparently disordered regions also undergo what is known as conditional folding. They remain flexible in isolation but adopt more defined structures when interacting with other molecules or when placed in specific cellular environments. AlphaFold occasionally captures hints of this behaviour.[PMC]pmc.ncbi.nlm.nih.govAlphaFold‐pLDDT uses 1 ‐ pLDDT as output.Read more…

That creates an interesting scientific opportunity. Instead of treating every low-confidence prediction as useless, researchers can use AlphaFold outputs as hypotheses about where disorder exists, where conditional structure might emerge, and where laboratory investigation should focus next.

In this sense, the system still accelerates research. It does not solve the disorder problem, but it helps scientists navigate it more efficiently.

Why this matters for the AI bloom debate

The broader significance extends beyond protein science.

AlphaFold is frequently cited as evidence that AI could accelerate discovery, improve medicine and contribute to a future of greater health and abundance. The achievement genuinely supports that possibility. A major scientific bottleneck became substantially easier to navigate, and researchers worldwide gained access to structural predictions that would previously have required immense resources.[EMBL]embl.orgalphafold potential impactsGreat expectations – the potential impacts of AlphaFold DBJul 22, 2021 — A discussion of the applications that AlphaFold DB may enabl…

Yet intrinsically disordered proteins reveal why scientific acceleration should not be confused with scientific completion.

Many of the hardest biological problems involve systems that are dynamic, probabilistic and context-dependent. Ageing, neurodegeneration, cancer progression and cellular regulation often depend on interactions across many molecules and many timescales. Predicting a single structure is only one piece of that puzzle.

The lesson is not that AI failed. The lesson is that real biological understanding often sits several layers above structure prediction. AlphaFold removed an enormous obstacle, but it did not eliminate the need for experiments, mechanistic theories, longitudinal studies or clinical evidence.

For advocates of an AI-enabled human bloom, this distinction matters. Progress in science may come less from a single dramatic breakthrough than from the cumulative removal of thousands of bottlenecks. AlphaFold showed that AI can remove one of them. Intrinsically disordered proteins remind us that many others remain.

The next frontier: from structures to ensembles

The challenge posed by disorder is already influencing the next generation of computational biology.

Researchers are increasingly exploring methods that predict structural ensembles rather than single structures. Instead of asking what shape a protein has, these approaches ask what range of shapes it can adopt and how often each state occurs.[Nature]nature.comWe benchmarked AF2…Read more…[Nature]nature.comHighly accurate protein structure prediction with AlphaFoldby J Jumper · 2021 · Cited by 50273 — The AlphaFold network directly pre…

That shift reflects a deeper change in how biology is modelled. Living systems are not collections of static molecular sculptures. They are networks of interacting molecules moving through fluctuating states.

If future AI systems become dramatically better at representing those dynamics, the implications could be substantial. Better modelling of disordered proteins might improve understanding of neurodegenerative disease, cellular signalling, aggregation disorders and ageing-related dysfunction. But reaching that point likely requires moving beyond the idea that biology can always be compressed into a single predicted structure.

Intrinsically disordered proteins therefore serve as a useful stress test for the entire AI-for-biology project. They mark the boundary where today’s remarkable successes encounter some of life’s most stubborn complexity. The fact that this boundary still exists does not weaken the case for AI-driven scientific acceleration. It clarifies what remains to be solved.

Amazon book picks

Further Reading

Books and field guides related to The proteins that do not hold still. Use these as the next step if you want deeper reading beyond the article.

Book

Life's Edge

By Carl Zimmer

Best broad fit for explaining living systems as dynamic and hard to reduce to fixed structures.

See on Amazon

Book

The Song of the Cell

By Siddhartha Mukherjee

Shows why proteins operate inside complex cellular environments rather than as isolated objects.

See on Amazon

Book

The Gene

By Siddhartha Mukherjee

Provides biological context for regulation, signalling and disease mechanisms involving proteins.

See on Amazon

Book

Transformer

By Nick Lane

Explains metabolism and molecular dynamism relevant to proteins that do not behave like fixed machines.

See on Amazon

Browse more on Amazon:Life's EdgeThe Song of the CellThe Gene

As an Amazon Associate I earn from qualifying purchases.

eBay marketplace picks

Marketplace Samples

Example marketplace items related to this page. Use the search link to explore similar finds on eBay.

Shop location

UsingUSA USA USAUKCanadaAustraliaIreland

Example eBay listing

Lysozyme protein structure Framed A Framed Wall Art Poster Canvas Print Picture

SearcheBay.co.uk: protein structure poster

Browse similar oneBay.co.uk

Example eBay listing

Lysozyme protein structure Framed A Framed Wall Art Poster Canvas Print Picture

SearcheBay.co.uk: protein structure poster

Browse similar oneBay.co.uk

Browse more oneBay.co.uk

Example items shown for inspiration; availability and pricing can change. Branchoria may earn a commission if you purchase through outbound eBay links.

Endnotes

1. Source: pmc.ncbi.nlm.nih.gov
Link:https://pmc.ncbi.nlm.nih.gov/articles/PMC10782001/

2. Source: nature.com
Link:https://www.nature.com/articles/s41467-026-69172-y

3. Source: arxiv.org
Link:https://arxiv.org/abs/2412.19875

4. Source: nature.com
Link:https://www.nature.com/articles/s41594-022-00849-w

5. Source: sciencedirect.com
Link:https://www.sciencedirect.com/science/article/abs/pii/S0022283621002771

6. Source: pmc.ncbi.nlm.nih.gov
Link:https://pmc.ncbi.nlm.nih.gov/articles/PMC10622901/

7. Source: nature.com
Link:https://www.nature.com/articles/s41586-021-03819-2

8. Source: nature.com
Link:https://www.nature.com/articles/s41586-021-03828-1

9. Source: sciencedirect.com
Link:https://www.sciencedirect.com/science/article/pii/S0022283621004411

10. Source: ebi.ac.uk
Title: strengths and limitations of alphafold
Link:https://www.ebi.ac.uk/training/online/courses/alphafold/an-introductory-guide-to-its-strengths-and-limitations/strengths-and-limitations-of-alphafold/

11. Source: pmc.ncbi.nlm.nih.gov
Link:https://pmc.ncbi.nlm.nih.gov/articles/PMC12956938/

12. Source: link.springer.com
Link:https://link.springer.com/article/10.1007/s10930-025-10310-8

13. Source: nature.com
Link:https://www.nature.com/articles/s41592-023-02087-4

14. Source: pmc.ncbi.nlm.nih.gov
Link:https://pmc.ncbi.nlm.nih.gov/articles/PMC9104326/

15. Source: pmc.ncbi.nlm.nih.gov
Link:https://pmc.ncbi.nlm.nih.gov/articles/PMC9601767/

16. Source: ebi.ac.uk
Link:https://www.ebi.ac.uk/training/online/courses/alphafold/[validation

17. Source: embl.org
Title: alphafold potential impacts
Link:https://www.embl.org/news/science/alphafold-potential-impacts/

18. Source: nature.com
Link:https://www.nature.com/articles/s41467-025-56572-9

19. Source: arxiv.org
Title: arXiv Deep Learning of Proteins with Local and Global Regions of Disorder
Link:https://arxiv.org/abs/2502.11326

20. Source: embl.org
Link:https://www.embl.org/

21. Source: embl.org
Link:https://www.embl.org/news/science-technology/first-complexes-alphafold-database/

22. Source: embl.org
Title: thornton alphafold
Link:https://www.embl.org/news/science/thornton-alphafold/

23. Source: nature.com
Link:https://www.nature.com/articles/s41392-023-01381-z

24. Source: nature.com
Link:https://www.nature.com/articles/s41586-024-07487-w

25. Source: nature.com
Link:https://www.nature.com/articles/s41467-023-44288-7

26. Source: sciencedirect.com
Link:https://www.sciencedirect.com/science/article/pii/S0022283625000336

27. Source: ebi.ac.uk
Link:https://www.ebi.ac.uk/training/online/courses/alphafold/inputs-and-outputs/evaluating-alphafolds-predicted-structures-using-confidence-scores/plddt-understanding-local-confidence/

Additional References

28. Source: alphafoldserver.com
Link:https://alphafoldserver.com/

29. Source: ebi.ac.uk
Link:https://www.ebi.ac.uk/training/online/courses/alphafold/inputs-and-outputs/evaluating-alphafolds-predicted-structures-using-confidence-scores/confidence-scores-in-alphafold-multimer/

30. Source: deepmind.google
Link:https://deepmind.google/science/alphafold/

31. Source: github.com
Link:https://github.com/google-deepmind/alphafold

32. Source: alphafold.ebi.ac.uk
Link:https://alphafold.ebi.ac.uk/

33. Source: youtube.com
Link:https://www.youtube.com/watch?v=5k8Lm9w9mpY

34. Source: youtube.com
Link:https://www.youtube.com/watch?v=UqeQfRDA8Yk

35. Source: youtube.com
Link:https://www.youtube.com/watch?v=i_aou7FRySw

36. Source: profiles.wustl.edu
Title: alphafold and implications for intrinsically disordered proteins
Link:https://profiles.wustl.edu/en/publications/alphafold-and-implications-for-intrinsically-disordered-proteins

37. Source: blog.google
Title: [Alpha Fold]({{ ‘alpha-fold/’ | relative_url }}) 3 predicts the structure and interactions of all
Link:https://blog.google/innovation-and-ai/products/google-deepmind-isomorphic-alphafold-3-ai-model/

Published: May 8, 2024