Why predicted binding is not a medicine

Introduction

AlphaFold 3 is one of the clearest examples of how AI might accelerate scientific discovery without eliminating the need for science itself. The system extends earlier AlphaFold models by predicting not only protein shapes, but also how proteins may interact with DNA, RNA, ions and small molecules, including many potential drug compounds. That brings AI closer to one of the most expensive and uncertain stages of medicine: finding molecules that might become useful treatments.[Nature]nature.comAccurate structure prediction of biomolecular interactions…by J Abramson · 2024 · Cited by 14929 — Here we describe our AlphaFol…

For supporters of an AI-enabled future of faster discovery, this matters because drug development is often slowed by the difficulty of understanding molecular interactions. Yet the key reality remains unchanged: a predicted interaction is not a medicine. A model can suggest that a drug candidate might bind to a target, but it cannot prove that the compound will work in living organisms, remain safe, reach the right tissues, avoid side effects, survive metabolism, or improve patient outcomes. AlphaFold 3 may help narrow the search space, but biology still decides what succeeds. Nature[EMBL-EBI]ebi.ac.ukEMBL-EBIHow have AlphaFold 3's predictions been validated?A critical finding from FoldBench is that AF3's ligand docking accuracy notably…

What AlphaFold 3 adds beyond protein shapes

The original AlphaFold breakthrough largely focused on predicting the three-dimensional structures of proteins. That was already valuable because protein shape strongly influences biological function. AlphaFold 3 moves into a more ambitious domain: modelling molecular interactions.

Instead of predicting a protein in isolation, the system can generate joint structures involving multiple molecular components. These may include proteins bound to DNA, RNA, metal ions, antibodies, modified biological molecules and small-molecule ligands, the category that contains many drugs.[Nature]nature.comprotein binder design and conformational state predictionby LA Abriata · 2026 — In the case of DeepMind's AlphaFold 2, the defining momen…[Isomorphic Labs]isomorphiclabs.comAlphaFold 3 predicts the structure and interactions of all of…May 8, 2024 — It models large biomolecules such as proteins, DNA, and RN…Published: May 8, 2024

This shift matters because most medicines work through interactions rather than static structures. A cancer drug, antibiotic or antiviral treatment usually succeeds only if it binds to a biological target in a particular way. Drug discovery therefore depends heavily on understanding questions such as:

AlphaFold 3 does not answer all of those questions, but it can provide plausible structural hypotheses much earlier in the research process. DeepMind reported major improvements over previous methods for several categories of molecular interaction prediction, including protein-ligand systems relevant to drug design.[blog.google]blog.googlegoogle deepmind isomorphic alphafold 3 ai modelAlphaFold 3 predicts the structure and interactions of all…8 May 2024 — Our new AI model AlphaFold 3 can predict the structure and int…Published: May 2024

In practical terms, this can help researchers decide which experiments are worth running. Instead of testing huge numbers of possibilities almost blindly, scientists may begin with AI-generated models that suggest promising directions.

That is the part of the AI bloom argument that deserves attention. Scientific acceleration does not necessarily require AI to solve biology outright. It may be enough for AI to reduce years of exploratory work into weeks or months, allowing researchers to spend more time validating ideas and less time searching for them.

Why predicted binding still fails

The public discussion around AlphaFold often creates an intuitive but misleading picture: if an AI predicts that a drug fits into a protein, then the treatment is halfway finished.

Drug developers know the reality is far harsher.

Binding is only one requirement

A molecule can bind beautifully in a computational model and still fail as a medicine.

Many drug candidates collapse because of problems that have little to do with binding geometry:

These problems routinely eliminate compounds that looked promising in early research. The history of pharmaceutical development is filled with molecules that appeared attractive in laboratory studies but failed in animals or human trials.

AlphaFold 3 can help with structural questions, but it does not directly solve pharmacology, toxicology, manufacturing or clinical effectiveness.[PMC]nih.govPMC7614146PMCAlphafold2 protein structure prediction : Implications for drug …by N Borkakoti · 2023 · Cited by 150 — Here we present our perspect…

Real proteins are moving targets

Another challenge is that proteins are not rigid objects.

Inside cells, proteins constantly shift between different conformations. Their behaviour can depend on temperature, chemical modifications, nearby molecules and cellular context. A predicted structure may capture one important state while missing others that matter for disease or treatment response.[The Guardian]theguardian.comThis breakthrough is expected to accelerate research in numerous fields, including antibiotics, cancer therapy, and agriculture. AlphaFol…

This matters because many drugs exploit dynamic behaviour rather than static shape. A compound may need to stabilise a rare protein state, block a temporary interaction or influence a changing molecular complex.

A prediction that appears accurate in a snapshot may therefore be incomplete from a therapeutic perspective.

Similar molecules are easier than novel ones

Independent evaluations have highlighted another important limitation. AlphaFold 3’s ligand predictions tend to become less reliable when the molecule being studied differs substantially from examples represented in its training data. Performance can drop when researchers move toward genuinely novel chemical space.[EMBL-EBI]ebi.ac.ukEMBL-EBIHow have AlphaFold 3's predictions been validated?A critical finding from FoldBench is that AF3's ligand docking accuracy notably…

That limitation matters because breakthrough medicines often involve new molecular designs rather than familiar ones.

An AI system can appear highly capable on benchmark problems yet become less dependable precisely where researchers most want creativity and generalisation.

Complex biological systems create cascading uncertainty

A successful drug does not interact with only one molecule.

Cells contain dense networks of proteins, nucleic acids, signalling pathways and feedback loops. A binding event that looks useful at the molecular level may have unexpected consequences when embedded inside a living organism.

This is one reason why drug development remains so expensive despite decades of advances in molecular biology. Understanding a single interaction is difficult. Understanding an entire biological system is vastly harder.

Even DeepMind’s researchers have emphasised that these problems are not fully solved and that substantial accuracy improvements remain necessary.[Axios]axios.comGoogle Deep Mind's new AI predicts how the molecules of life interactBuilding on the success of the original AlphaFold, which solved the complex problem of predicting protein structures from amino acid sequ…

Where prediction fits in the drug pipeline

The most realistic way to think about AlphaFold 3 is not as a replacement for drug discovery but as a tool within it.

Drug development typically moves through several stages:

AlphaFold 3 mainly affects the earliest stages.

Its greatest value may be helping researchers:

Researchers have already shown examples where AlphaFold-based approaches helped accelerate early-stage hit discovery against difficult targets. However, even these success stories still required chemical synthesis, biological testing and repeated experimental validation.[arXiv]arxiv.orgAlphaFold Accelerates Artificial Intelligence Powered Drug Discovery: Efficient Discovery of a Novel Cyclin-dependent Kinase 20 (CDK…

The distinction matters because public discussion often compresses the entire drug pipeline into a single step. In reality, AlphaFold operates near the beginning of a long chain.

A useful analogy is that it can help produce better maps. Better maps can make exploration faster and cheaper, but explorers still have to travel the terrain.

The strongest case for scientific acceleration

The most persuasive argument linking AlphaFold 3 to a larger future of AI-enabled abundance is not that it instantly cures disease.

It is that scientific research contains many bottlenecks that are fundamentally information problems.

Researchers often spend enormous amounts of time asking questions such as:

If AI systems become increasingly capable of narrowing those uncertainties, the cumulative effect across biology could be substantial.

Instead of reducing a ten-year drug development process to a few days, AI may reduce thousands of dead ends throughout the research ecosystem. The resulting gains could compound across universities, biotech firms, hospitals and public-health institutions.

In the broader AI bloom vision, this is important because long-run progress may come less from a single miraculous breakthrough than from repeated reductions in scientific friction. Faster hypothesis generation, improved modelling and better experimental targeting could gradually increase the rate at which humanity discovers treatments, understands disease and extends healthy life. AlphaFold provides one of the clearest demonstrations that such acceleration is possible.[Google DeepMind]blog.googlegoogle deepmind isomorphic alphafold 3 ai modelAlphaFold 3 predicts the structure and interactions of all…8 May 2024 — Our new AI model AlphaFold 3 can predict the structure and int…Published: May 2024[Nature]nature.comAccurate structure prediction of biomolecular interactions…by J Abramson · 2024 · Cited by 14929 — Here we describe our AlphaFol…

Why laboratory and clinical proof still matter

The most important reality check is that medicine ultimately concerns people, not molecular models.

A prediction can suggest that a compound binds to a target. It cannot demonstrate that patients live longer, suffer less pain, avoid disability or experience acceptable safety outcomes.

Those questions require experiments because living organisms contain layers of complexity that current models only partially capture.

Independent benchmarking has also found important weaknesses in AlphaFold 3, including reduced reliability in flexible molecular regions, difficulty with some antibody-antigen interactions and performance declines under certain challenging conditions.[EMBL-EBI]ebi.ac.ukEMBL-EBIHow have AlphaFold 3's predictions been validated?A critical finding from FoldBench is that AF3's ligand docking accuracy notably…

That does not diminish the achievement. It simply clarifies what the achievement actually is.

AlphaFold 3 represents a major advance in predicting biological interactions. It may help researchers move more quickly from question to hypothesis and from hypothesis to experiment. But drug discovery remains a process of evidence, not prediction alone.

For anyone thinking about AI and humanity’s long-term future, that distinction is worth remembering. The strongest path to medical abundance is unlikely to come from AI replacing science. It is more likely to come from AI making science dramatically faster, while experiments, clinical trials and real-world evidence continue to determine what is true. Nature[EMBL-EBI]ebi.ac.ukEMBL-EBIHow have AlphaFold 3's predictions been validated?A critical finding from FoldBench is that AF3's ligand docking accuracy notably…

Amazon book picks

Further Reading

Books and field guides related to Why predicted binding is not a medicine. Use these as the next step if you want deeper reading beyond the article.

Book

The Drug Hunters

By Donald R. Kirsch, Ogi Ogas

Explains why predicted binding is only an early step toward a real drug.

See on Amazon

Book

Deep Medicine

By Eric Topol

Frames AI prediction as useful only when connected to evidence and patient outcomes.

See on Amazon

Book

The Code Breaker

By Walter Isaacson

Gives context for AI-enabled molecular biology and the difference between discovery tools and cures.

See on Amazon

Book

The Billion-Dollar Molecule

By Barry Werth

Shows how promising molecular ideas encounter uncertainty in real pharmaceutical development.

See on Amazon

Browse more on Amazon:The Drug HuntersDeep MedicineThe Code Breaker

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

1920s/1930s Bakelite Molecule Model Set Made By Catalina, Waltham Abbey

SearcheBay.co.uk: molecule model

Browse similar oneBay.co.uk

Example eBay listing

Vintage Bohemian Molecule Scientific Model Sculpture 18cm Modernist Czech 1950s

SearcheBay.co.uk: molecule model

Browse similar oneBay.co.uk

Example eBay listing

Chemistry Molecule Model Building Kit

SearcheBay.co.uk: molecule model

Browse similar oneBay.co.uk

Example eBay listing

Molecular Model Kit (238pcs), Chemistry Molecule Science Gifts for Ochem1604

SearcheBay.co.uk: molecule model

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

2. Source: blog.google
Title: google deepmind isomorphic alphafold 3 ai model
Link:https://blog.google/innovation-and-ai/products/google-deepmind-isomorphic-alphafold-3-ai-model/

Published: May 2024

3. Source: ebi.ac.uk
Link:https://www.ebi.ac.uk/training/online/courses/alphafold/alphafold-3-and-alphafold-server/introducing-alphafold-3/how-have-alphafold-3s-predictions-been-validated/

4. Source: ebi.ac.uk
Title: introducing alphafold 3
Link:https://www.ebi.ac.uk/training/online/courses/alphafold/alphafold-3-and-alphafold-server/introducing-alphafold-3/

5. Source: axios.com
Title: Google Deep Mind’s new AI predicts how the molecules of life interact
Link:https://www.axios.com/2024/05/09/google-deepmind-ai-new-alphafold-3

6. Source: arxiv.org
Link:https://arxiv.org/abs/2201.09647

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

8. Source: arxiv.org
Link:https://arxiv.org/abs/2406.03979

9. Source: google.com
Link:https://www.google.com/

10. Source: nature.com
Link:https://www.nature.com/articles/s42003-026-10112-3

11. Source: about.google
Link:https://about.google/intl/ja_ALL/

12. Source: mdpi.com
Link:https://www.mdpi.com/2218-273X/14/3/339

13. Source: theguardian.com
Link:https://www.theguardian.com/science/article/2024/may/08/google-deepmind-ai-biology-alphafold

14. Source: isomorphiclabs.com
Link:https://www.isomorphiclabs.com/articles/alphafold-3-predicts-the-structure-and-interactions-of-all-of-lifes-molecules

Published: May 8, 2024

15. Source: isomorphiclabs.com
Title: the isomorphic labs drug design engine unlocks a new frontier
Link:https://www.isomorphiclabs.com/articles/the-isomorphic-labs-drug-design-engine-unlocks-a-new-frontier

16. Source: isomorphiclabs.com
Title: rational drug design with alphafold 3
Link:https://www.isomorphiclabs.com/articles/rational-drug-design-with-alphafold-3

Published: May 8, 2024

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

18. Source: reddit.com
Link:https://www.reddit.com/r/science/comments/1cn7le6/google_deepmind_alphafold_3_predicts_the/

Additional References

19. Source: medium.com
Link:https://medium.com/%40cognidownunder/alphafold-changed-biology-forever-when-it-solved-protein-folding-78bb8768483a

20. Source: linkedin.com
Link:https://www.linkedin.com/posts/nexco-analytics_the-limitations-of-protein-ligand-co-folding-activity-7396260189127118848-bQg9

21. Source: medium.com
Link:https://medium.com/%40sinclairhuang/alphafold-3-revolution-or-reality-check-what-industry-insiders-are-saying-e13991063f2a

22. Source: prescouter.com
Link:https://www.prescouter.com/2024/05/alphafold-3/

23. Source: sciencedirect.com
Link:https://www.sciencedirect.com/science/article/pii/S2590123026021237

24. Source: linkedin.com
Link:https://www.linkedin.com/pulse/impact-alphafold-3s-open-source-release-life-sciences-xinyu-zhang-wa5ue

25. Source: frontiersin.org
Link:https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2026.1739303/full

26. Source: drugdiscoverytrends.com
Link:https://www.drugdiscoverytrends.com/meet-alphafold-3-which-can-accurately-model-more-than-99-of-molecular-types-in-the-protein-data-bank/

Published: May 2024

27. Source: lemonde.fr
Link:https://www.lemonde.fr/en/science/article/2024/10/09/nobel-prize-for-chemistry-2024-artificial-intelligence-garners-more-recognition_6728828_10.html

28. Source: rcastoragev2.blob.core.windows.net
Link:https://rcastoragev2.blob.core.windows.net/9e6c827f9d7f88eac8af0797432cefc7/41586_2024_7487_MOESM1_ESM.pdf