AlphaFold models cover over 90% of the human NPC scaffold

Researchers combined AI predictions with electron-microscopy maps to model more than 90% of the human nuclear pore scaffold.

22ⓘ

AI system

System

AlphaFold, RoseTTAfold, and ColabFold. [1]

Model

AlphaFold and RoseTTAfold. [1]

Release

The 2021 research configuration. Exact builds remain unverified. [1] [2] [3] [4]

Provider

DeepMind (AlphaFold); University of Washington and Harvard University researchers (RoseTTAFold) [3] [4]

Role

The models predicted protein structures and complexes for comparison with microscopy data. [1]

Configuration

Researchers used AlphaFold and RoseTTAfold for protein structures, ColabFold for complexes, and Assembline to fit the models to microscopy data. [1]

Inputs and outputs

Protein sequences supplied the basis for predictions of individual structures and complexes. Researchers fitted those predictions to electron-microscopy maps. [1]

Autonomy

Not assessed

Workflow

Cryo-electron tomographic maps of the human nuclear pore complex in two states. AI structure predictions for nucleoporins and subcomplexes, benchmarked against experimental structures. Integrative modeling by groups holding prior structural data on the complex.

Human oversight

Not assessed

Relation to AI

AI structure prediction generated the nucleoporin and subcomplex models that formed the input to integrative modeling. The authors state that the predictions were benchmarked against existing and new experimental structures. The resulting model was validated against earlier and new experimental data, so the AI output was independently checked rather than accepted as given.

AI contribution

AI structure prediction generated the nucleoporin and subcomplex models that formed the input to integrative modeling. The authors state that the predictions were benchmarked against existing and new experimental structures. The resulting model was validated against earlier and new experimental data, so the AI output was independently checked rather than accepted as given.

Benefits

Outcome identifiers

Not assessed

Benefit

covers more than 90% of the NPC scaffold, roughly doubling previous structural coverage

Category

Improved Knowledge Discovery

Status

Not assessed

Beneficiary groups

Organizations. Society.

Beneficiaries

Not assessed

Reach

Not assessed

Distribution and access

Not assessed

Duration

Not assessed

Expected?

Yes

Costs and harms

Not assessed

Measurement

Studies

Not assessed

Study design

Not assessed

Measurement

Not assessed

Setting and sample

Not assessed

Measures

covers more than 90% of the NPC scaffold, roughly doubling previous structural coverage

Comparison

the previous structural model of the human NPC scaffold

Results

Reference: 16 Nups and 35 MDa (46% of the scaffold) of the previous model. Result: 25 out of ~30 Nups, and account for 70 MDa of the molecular weight of the NPC (>90% of the scaffold molecular weight).

Additional value from AI

Not assessed

Limitations

Not assessed

Denominators

Not assessed

Materials for replication

Not assessed

Money

Kind of value

Not assessed

Amount

Not assessed

Currency and prices

Not assessed

Period and unit

Not assessed

Realization

Not assessed

Recipients and payers

Not assessed

Gross or net

Not assessed

Included costs

Not assessed

Comparison

Not assessed

Method

Not assessed

Return and payback

Not assessed

People and places

Participants

DeepMind developed AlphaFold. Researchers at the University of Washington and Harvard developed RoseTTAFold. The EMBL and Max Planck groups used the software in this study. [1] [3] [4]

Jurisdictions

Benefit: Heidelberg, Frankfurt am Main and Hamburg, Germany. Operator: Germany.

Ownership and control

Not assessed

Dependencies

Not assessed

Cross-border connections

Not assessed

Geopolitical implications

Not assessed

Follow-up

Conditions for replication

Cryo-electron tomographic maps of sufficient resolution for fitting. Nucleoporin sequences and homologues available for prediction. Independent experimental structures available for benchmarking.

Replication attempts

Not assessed

Later findings

No later development found.

Availability

Not assessed

Institutional decision

Not assessed

Grounds

Not assessed

Action and result

Not assessed

Success criterion

Not assessed

Evidence

Source numbers

AIBD-0043-source-0001–0004

Citations

Numbered references below.

Language

Not assessed

Source relationships

Operator, Developer

Original evidence

The authors’ preprint identifies the models and their results. Original developer records identify the model providers. [1] [2] [3] [4]

Access and versions

An author-posted copy of the November 2021 preprint was accessible. The final Science article and its supplementary methods were unavailable. [1] [2] [3] [4]

Supported findings

Not assessed

Evidence strength

Not assessed

Missing evidence

Not assessed

Grades

Assessed benefits

Organizations. Society.

Depth

Organizations: 2. Society: 2.

Reasons

Organizations: The collaborating laboratories obtained a model that roughly doubles structural coverage of the human NPC scaffold. That is a material advance for those groups' research programs. Society: A 70-MDa atomically resolved model covering over 90% of the scaffold is a durable public resource for nuclear transport research. The source reports no direct patient or population benefit.

Observed or intended

Not assessed

Rubric

Not assessed

Assessor and date

Not assessed

Alternative assessments

Not assessed

Fill

The deployer's account

Entry details

Kind

Not assessed

Field

Structural and cell biology.

Purpose

Predict protein and subcomplex structures to enable integrative modeling of a 120-MDa macromolecular machine that resisted conventional structural biology.

Dates

Source published: 27 October 2021. Updated: 25 October 2026.

Stage

Not assessed

Version

SHA-256: b7d515a86a02…

References

Other identifiers

Not assessed

Category mappings

Not assessed

Brief

Not assessed

Revisions

6 revisions. Latest: 25 October 2026.

Sources

  1. Preprint Deployer

    Mosalaganti S, Obarska-Kosinska A, Siggel M, et al. (2021). Artificial intelligence reveals nuclear pore complexity. bioRxiv 2021.10.26.465776, posted 27 October 2021.

    Original manuscript identifies the structure-prediction tools and the coverage of the nuclear pore scaffold. Introduction; Results, “A near-complete model of the human NPC scaffold.” Access copy

    Original manuscript from the same study as the published Science article. The inspected access copy is the version posted November 2, 2021.

  2. Peer-reviewed study Deployer

    Mosalaganti S, Obarska-Kosinska A, Siggel M, et al. (2022). AI-based structure prediction empowers integrative structural analysis of human nuclear pores. Science.

    Published version of the same research. Its full text was not accessible during this review. Abstract; bibliographic record.

    Published version of the same study as the preprint.

  3. Report Developer

    Google DeepMind. “AlphaFold.” Timeline of a breakthrough, entries for 2020 and 2021.

    Original developer account identifies DeepMind as AlphaFold’s developer and documents the 2021 release chronology. Timeline of a breakthrough, entries dated 30 November 2020 and 15 July 2021.

    Developer documentation. It does not independently evaluate the nuclear-pore work.

  4. Other Developer

    Baek, Minkyung, Frank DiMaio, Ivan Anishchenko, Justas Dauparas, Sergey Ovchinnikov, Jue Wang, and David Baker. “RoseTTAFold: The First Release of RoseTTAFold.” Version 1.0.0. Zenodo, July 3, 2021. doi:10.5281/zenodo.5068265.

    The original RoseTTAFold release record names the authors and their University of Washington and Harvard affiliations. Authors/Creators; affiliations; description.

    Software release deposited by its developers. Its version number does not establish the exact build used in this study.