Structure - Methods summary

Summary

The Structure resource contains information about the three-dimensional structure, post-translational modifications (PTMs) and genetic variants of human proteins. Interactive 3D protein structures based on predictions generated using the AlphaFold source code are shown for most human proteins and their related isoforms. The Protein Browser tool can be used to select among the different isoforms and display protein related features such as known antigen sequences, PTMs, transmembrane regions and InterPro domains on the structures. The amino acid positions of population and clinical genetic variants can also be displayed.

Key publications

Jumper J et al. (2021) "Highly accurate protein structure prediction with AlphaFold" Nature 596(7873):583-589.

Varadi M et al. (2022) "AlphaFold Protein Structure Database: massively expanding the structural coverage of protein-sequence space with high-accuracy models" Nucleic Acids Research 50(D1):D439-D444.

Han Y et al. (2025) "MTPrompt-PTM: A Multi-Task Method for Post-Translational Modification Prediction Using Prompt Tuning on a Structure-Aware Protein Language Model" Biomolecules 15(6):843.

Chung CR et al. (2025) "dbPTM 2025 update: comprehensive integration of PTMs and proteomic data for advanced insights into cancer research" Nucleic Acids Res 53(D1):D377-D386.

Cheng J et al. (2023) "Accurate proteome-wide missense variant effect prediction with AlphaMissense" Science Vol 381, Issue 6664.

What can you learn from the Structure resource?

Learn about:

  • the predicted 3D structure of proteins and their related isoforms
  • the antigen structure for the majority of the antibodies
  • the predicted structure of selected protein features
  • the known and predicted PTMs for the proteins
  • the known and predicted missense variants with clinical significance
  • the known population variants and predicted benign missense variants

How has the data been generated?

Structures

The predicted 3D protein structures have been generated in-house based on the AlphaFold source code developed by Deepmind which was modified for running on split CPU/GPU clusters. The structure predictions were run on National Supercomputer Centre (NSC) at Linköping University.

The AI-system Alphafold is a machine learning approach in which the primary amino acid sequence and aligned sequences of homologues together with physical and biological knowledge about related protein structures are incorporated into the design of a deep learning algorithm to directly predict the 3D structure of a protein. Structure predictions have been made for all protein isoforms that contain exclusively standard amino acids and have a length between 6 aa and 4000 aa. Secondary structures were assigned using a reimplementation of the DSSP algorithm from OpenStructure (Biasini M et al. (2013))

PTMs

Experimental and predicted data was obtained for six common types of PTMs, phosphorylations, N-glycosylation, methylations, acetylations, sumoylations and ubiquitinations.

The multi-task deep learning framework MTPrompt_PTM version 4428d03 was used to predict PTMs for all protein isoforms containing exclusively standard amino acids and with a length shorter than 1022 aa based on the developers recommendation. Predictions above our selected cut-off >0.95 were obtained for 69392 isoforms corresponding to 18061 genes. The predicted PTMs are displayed in the 3D structures as well as in the PTM Protein Browser feature.

Experimental PTMs were obtained from UniProt version 2026_03 and dbPTM version 2025. The PTMs downloaded from dbPTM was mapped to ensembl isoforms based on 100% identity to sequences corresponding to the UniProt identifier obtained in the file, and having the same PTM position and surrounding sequence as presented in dbPTM. This resulted in dbPTM data for 37442 protein isoforms corresponding to 18093 genes and UniProt data for 17147 protein isoforms corresponding to 7936 genes.

Genetic variants

Missense variant data is included from two sources. The known population and clinical variants data is incorporated from the Ensembl Variation database, with clinical relevance being based on the clinical significance terms "pathogenic" and "likely pathogenic". The predicted benign and pathological missense variants are integrated from AlphaMissense with "likely benign" variants being defined as having a pathogenicity score <0.05 and "likely pathogenic" as having a pathogenicity score >0.95.

All structures are displayed using the NGL Viewer.


What is presented in the resource?

In the gene summary page predicted 3D protein structures can be displayed and explored. The structures for the different protein isoforms can be selected and displayed by clicking on the transcript names on top of the ProteinBrowser tool, next to which the schematic structure of the different splice variants including exons, introns and UTRs are displayed. By clicking on e.g. antigens, PTMs from the three different datasets, InterPro domains or membrane regions in the schematic protein view in the ProteinBrowser, these features will be highlighted in the 3D structure. The sliders beside the structure allow highlighting of predicted or experimentally determined positions for the six types of PTMs, positions of missense variants as well as coloring of the structure according to the AlphaFold confidence measure or residue index.