GTalign web server help
Introduction
The GTalign web server (GTalign-web) is designed to harness the capabilities of the GTalign protein structure alignment and search tool [1]. Powered by spatial structure indexing, GTalign achieves high accuracy by thoroughly exploring the superposition space while maintaining high speed.
GTalign-web is engineered to handle large-scale structure alignment and search tasks. It supports both single-query and batch alignment modes, allowing users to input multiple structures in diverse formats. Pairwise alignments resulting from searches against large structure databases can be analyzed using interactive visualization tools. GTalign-web also provides functionality for constructing multiple structure alignments based on identified hits.
Input
GTalign-web allows users to input single or multiple structure queries in various formats, with submissions requiring a single file. The maximum file size for upload is 10 MB.
For a single query, the input file can be optionally gzipped and must be in either PDB or PDBx/mmCIF format with appropriate extensions. Supported extensions are:
- PDB: .pdb, .ent, .pdb.gz, .ent.gz
- PDBx/mmCIF: .cif, .cif.gz
For multiple queries, the same format and extension rules apply.
Additionally, multiple files may be archived (without compression) into a single .tar file using the tar utility. For example, to submit two queries named my_structure1.pdb.gz and my_structure2.cif.gz, archive them, e.g., using the command:
tar -cf my_structures.tar my_structure1.pdb.gz my_structure2.cif.gz
The resulting file, my_structures.tar, can then be uploaded. The .tar extension is mandatory.
Results will be generated for each query individually. If any of the structure files, whether single or multiple queries, contain multiple chains and/or models, each chain and model will be treated as a separate query and processed independently for results.
Queries shorter than 20 residues or longer than 4000 residues will be ignored.
Databases
GTalign-web provides access to the following structure databases for search:
- PDB: RCSB Protein Data Bank [2]
- SCOPe40: Structural Classification of Proteins (extended), filtered to 40% sequence identity [3]
- ECOD-F70: Evolutionary Classification of protein Domains [4]
- UniProtKB/Swiss-Prot: Reviewed entries from the UniProt Knowledgebase [5,6]
- Reference proteomes: A collection of 48 representative proteomes [5,6]
- UniRef30: UniProt entries clustered at 30% sequence identity [5,6]
- BFVD: the Big Fantastic Virus Database [7]
Search options
GTalign-web offers a variety of customizable options to control the search process:
- TM-score threshold: Users can set a minimum TM-score threshold (default = 0.5). Alignments with lower TM-scores will not be shown.
- Sorting options: Results can be sorted using the following criteria:
- 0: Greater TM-score of the two structures
- 1: Reference length-normalized TM-score
- 2: Query length-normalized TM-score (default)
- 3: Harmonic mean of the two TM-scores
- 4: RMSD
- 5: Greater 2TM-score
- 6: Reference length-normalized 2TM-score
- 7: Query length-normalized 2TM-score
- 8: Harmonic mean of the two 2TM-scores
- Number of alignments: The number of results displayed is capped at 1,000.
- Similarity pre-screening:
- Pairwise sequence similarity score: Filter structures based on minimum sequence similarity (default = 0 / not used; maximum = 100).
- Provisional structural TM-score: Filter structures based on a minimum TM-score (default = 0.4).
- Search speed optimization: Users can adjust the GTalign algorithm to balance speed and alignment optimality. Larger values result in faster computations but reduced accuracy. The default value of 13 corresponds to the fastest setting, ideal for large databases like UniRef30. Note that searches using the most accurate setting (value 0) may take up to 10 times longer compared to the fastest setting.
- Gaps removal: Users can opt to exclude deletion positions (gaps in the query) from alignments, which is particularly useful for evolutionary analyses when constructing multiple structure alignments.
Description of results and services
The entry point to GTalign-web's services is job submission, where users search selected protein structure databases for structural similarities. Once the computations are completed, GTalign-web displays a list of links to the results for each query structure. Clicking a link directs the user to a detailed results page for the individual query (Fig. 1).
The results page provides a list of GTalign-generated pairwise structure alignments sorted by structural similarity (Fig. 2). Each identified structure is linked to its original database entry. Each alignment represents a match between the query and a database entry, including aligned secondary structure sections to help assess topological similarity. Secondary structure elements are denoted by the following letters:
- h: alpha helix
- e: beta strand
- t: turn
The + symbol in the middle line indicates pairs of aligned residues within 5 angstroms.
Each structure alignment is accompanied by the following statistics:
- TM-score (Refn./Query): Length-normalized TM-scores for the reference (database entry) and the query [8]. A TM-score of 1 indicates a perfect match.
- 2TM-score: Secondary TM-score that excludes unmatched helices [1]. A 2TM-score of 1 also indicates a perfect match.
- d0 (Refn./Query): Normalizing inter-residue distance (d0) for the reference and query [8].
- RMSD: Root-mean-square deviation in angstroms (A).
- Identities, Matched, and Gaps: These indicate the number of identical aligned residues, residues aligned within 5 angstroms, and unaligned residues, respectively. Their percentages sum to 100%.
Rotation matrices and translation vectors (transformation matrices) for superimposing the reference structure onto the query structure are also provided. GTalign-web enables interactive visual analysis of superimposed structures.
Additionally, GTalign-web offers functionality for constructing multiple structure alignments based on selected pairwise alignments. Users can either select alignments individually or specify a TM-score range to filter alignments. The resulting multiple structure alignments are presented in the form of a multiple sequence alignment and are compatible with interactive visual analysis tools.
A brief guide to TM-scores
GTalign-web ranks alignments using several measures, with the most widely used being the TM-score. The TM-score is a global measure of structural similarity between two protein structures, which can be normalized by the length of either the query or the reference (subject) structure.
The TM-score is intuitive: it can be loosely interpreted as the fraction of a protein that aligns perfectly with another. For example, a TM-score of 0.5 suggests that approximately half of one protein aligns closely with the other.
In practice, however, structural alignments typically include insertions and deletions, so perfect alignment is rare. In these cases, the TM-score reflects how closely one protein's structure fits the other in three-dimensional space, with higher TM-scores indicating closer spatial alignment. Empirically, a TM-score of 0.5 or greater is a strong indicator that two proteins share (with high probability) a common structural fold.
By default, GTalign-web ranks alignments using the TM-score normalized by the query length, so the significance of the TM-score should be interpreted in relation to the query structure. For example, a TM-score of 0.5 means that the query likely shares the same fold as the matched region in the reference.
Alternatively, the TM-score normalized by the shorter protein (i.e., the greater of the two TM-scores) reflects how well the shorter structure aligns within the context of the longer one.
For comparisons between proteins of similar lengths, sorting by the harmonic mean of the two normalized TM-scores is often appropriate.
Lastly, GTalign-web also provides the 2TM-score (secondary TM-score), which is calculated by excluding unmatched helices. This makes it a stricter measure for assessing topological similarity. All the above significance thresholds apply equally to the 2TM-score.
Runtimes
The table below shows the runtimes of GTalign searches against the PDB, UniProtKB/Swiss-Prot, and UniRef30 structure databases for query structures of varying lengths, using default server settings. These runtimes were measured using the server's API. A dozen of seconds would be required to render the results in the graphical user interface.
For shorter query structures (<500 residues) against the UniRef30 database, runtimes appear similar due to the bottleneck caused by slower data retrieval compared to computation. A potential solution to this issue is upgrading to faster storage systems, which will be considered in future updates.
## Runtime (s) ## -------------------------------------- ## Structure_ID Length PDB Swiss-Prot UniRef30 ## (981,249 chains) (541,741 strs.) (25,348,391 strs.) ## (avg_len: 216) (avg_len: 354) (avg_len: 223) ## A0A0H1RF20 100 39 29 1257 A0A7S2FVF0 100 39 30 1255 A0A7S2FSH7 100 42 30 1262 A0A357HZ95 200 41 32 1282 A0A225V1N4 200 38 26 1264 A0A7S2WTE7 200 39 29 1271 K1Z1K8 500 71 48 1802 A0A447Y1L5 500 64 43 1736 A0A6P0TGX7 500 69 47 1786 A0A2V9DNX9 1000 126 85 2845 A0A1I8J5K6 1000 120 84 2809 A0A524N3P2 1000 127 88 3058 A0A1C1D0Y4 2004 223 155 5517 A0A158Q3Q4 2005 213 146 5248 A0A175WFP4 2006 227 159 5760 7k17_A 3629 395 282 9952 8j07_g1 3939 415 278 10419 7k5b_C 3947 421 287 10692
Limits
- File size: The maximum upload size for input files is 10 MB.
- File name: The maximum length of the name of uploaded file is 90 characters.
- Residue length: Query structures shorter than 20 residues or longer than 4000 residues will not be processed.
- Job duration: The maximum runtime for a job is 24 hours. If a job exceeds this limit, it will be canceled, and the user will be notified.
- Jobs are saved for 2 weeks.
If some queries in a job fail, an error log will provide a brief description of the problem, visible to the user. Results for successfully completed queries within the same job will remain available for analysis as usual.
References
- Margelevicius M. GTalign: spatial index-driven protein structure alignment, superposition, and search. Nature Communications 15, 7305 (2024).
- Burley S et al. RCSB Protein Data Bank: powerful new tools for exploring 3D structures of biological macromolecules for basic and applied research and education in fundamental biology, biomedicine, biotechnology, bioengineering and energy sciences. Nucleic Acids Res. 49, D437-D451 (2021).
- Chandonia J, Fox N, Brenner S. SCOPe: classification of large macromolecular structures in the structural classification of proteins-extended database. Nucleic Acids Res. 47, D475-D481 (2019).
- Schaeffer R, Liao Y, Cheng H, Grishin N. ECOD: new developments in the evolutionary classification of domains. Nucleic Acids Res. 45, D296-D302 (2017).
- UniProt Consortium. UniProt: the universal protein knowledgebase in 2021. Nucleic Acids Res. 49, D480-D489 (2021).
- Varadi M. et al. AlphaFold Protein Structure Database in 2024: providing structure coverage for over 214 million protein sequences. Nucleic Acids Res 52, D368-D375 (2024).
- Kil RS et al. BFVD-a large repository of predicted viral protein structures. Nucleic Acids Res, 53, D340-D347 (2025).
- Zhang, Y. and Skolnick, J. Scoring function for automated assessment of protein structure template quality. Proteins 57, 702-710 (2004).