FoldSeek

FoldSeek webserver overview

FoldSeek searches, compares and clusters protein structures. Search public databases or your own target collection, compare two structures, find reciprocal hits, or analyze multichain complexes and interfaces. Local collections can contain up to 2,500 structures. Results include native alignments, scores and downloadable files.

Set up and inspect runs in your browser. Structures can be uploaded or fetched from the Protein Data Bank, and completed jobs remain available for inspection, download or reruns. Computation runs on hosted workers or the public FoldSeek search service. No local FoldSeek installation or database download is required.

Pricing

Jobs start at 50 credits, and larger structure inputs can cost more. Database search quotes each submitted structure separately; local comparisons, searches and clustering quote the selected structures together. Changing the native algorithm or output format does not itself add a price multiplier. The exact credit price is shown before submission and is also available through the API.

FoldSeek API

Workspaces on a paid plan can run the FoldSeek server from scripts, notebooks, and pipeline services through the ProteinIQ API. Start with GET /api/v1/tools/foldseek to retrieve the active input contract, available modes and settings, and current credit model. The same endpoint keeps integrations aligned with changes to supported inputs or limits.

Submit PDB or mmCIF structures in input.inputs[] and choose a mode listed below. Collection searches use the structures and targets slots. Request a quote through POST /api/v1/jobs/quote or create a job with POST /api/v1/jobs. Monitor the status or events endpoints before retrieving results and signed downloads. API tools documents tool contracts, and API jobs covers quoting, submission, status and results.

What is FoldSeek?

FoldSeek is a fast protein structure search tool. It can search selected AlphaFold Database datasets and other structural databases, compare structures in detail, or cluster structures by similarity. Database search uses the versions currently advertised by the public service.

Traditional structure comparison methods like TM-align are accurate but slow, requiring seconds per comparison. FoldSeek achieves comparable sensitivity while being four to five orders of magnitude faster. This speed comes from a novel encoding approach that converts 3D coordinates into searchable sequences.

For sequence-based clustering, use MMseqs2. For detailed pairwise structure alignment with superposition, use USAlign.

Database search

Upload a single structure to search against massive structure databases. Database search uses the FoldSeek web server API, giving you access to the same search capabilities as search.foldseek.com.

Available databases include AlphaFold DB 50, PDB, AlphaFold Swiss-Prot, CATH 50, BFVD, and AlphaFold Proteome. AlphaFold DB 50 and PDB are selected by default. Database search typically completes in 1-5 minutes depending on server load.

Every native query chain is retrieved. The Data tab and convenience TSV show up to 500 hits; the complete per-chain responses and service result archive remain downloadable. If all database switches are off, the maintained selection is AlphaFold DB 50 and PDB. The saved catalog records the database versions and search parameters reported by the service, including the BFVD v3 identifier.

How does FoldSeek work?

The 3Di structural alphabet

FoldSeek's speed comes from the 3Di (3D interaction) alphabet, which encodes protein structure as a sequence of 20 letters. Unlike traditional backbone structural alphabets, 3Di describes the geometric relationship between each residue and its spatially closest neighbor.

For each residue $i$, FoldSeek finds its nearest neighbor residue $j$ based on virtual center distance. Seven angles, the $C_\alpha$ distance, and two sequence distance features are extracted from the backbone coordinates of both residues. These 10 features define the 20 3Di states through a neural network trained to maximize evolutionary conservation.

This encoding has three advantages over backbone alphabets: weaker dependency between consecutive letters, more evenly distributed state frequencies, and higher information density in conserved protein cores rather than loop regions.

Search algorithm

FoldSeek converts both query and target structures into 3Di sequences. It then applies the MMseqs2 prefilter to find candidate matches using spaced k-mer matching on diagonals of the alignment matrix. This prefilter reduces the search space by several orders of magnitude while maintaining high sensitivity.

For hits passing the prefilter, FoldSeek performs Smith-Waterman local alignment combining both 3Di and amino acid substitution scores. The final alignment uses structural superposition to calculate TM-score and LDDT.

Inputs

FoldSeek accepts protein structures as uploaded files or RCSB PDB fetches. The selected mode determines how many structures belong in a job.

Input	Accepted formats	Limits and behavior
Protein structures	.pdb, .ent, .cif, .mmcif, or an RCSB PDB structure	Files can be up to 50 MiB. Database search accepts one structure, pairwise comparison requires exactly two, and clustering accepts 3-2,500 structures.
Gzip uploads	.pdb.gz, .ent.gz, .cif.gz, .mmcif.gz	The upload reader decompresses these to coordinate text before admission. The same size and structure limits apply. API submissions use the uncompressed coordinate representation, with UTF-8 or base64 transport.
Query and target collections	The same coordinate formats in separate input fields	Local, complex, interface and reciprocal-hit searches require at least one query and one target, with at most 2,500 structures in total.
Job name	Text	Optional label used to identify the run.

Database search creates one independent search job per submitted structure. Other modes keep each submitted collection together. Complex modes align whole multichain assemblies. Interface modes use Foldseek's own interface extraction and comparison. A structure count that does not match the selected mode is rejected.

Settings

The selected mode controls which settings apply. Blank optional controls preserve the native mode default, except for the maintained pairwise and monomer-clustering presets described below.

Parameter	Type	Default	Description
Mode	enum	database_search	Also supports pairwise, cluster, local_search, multimer_search, interface_search, reciprocal_search, multimer_cluster and interface_cluster.
Database switches	boolean	AlphaFold DB 50 and PDB on	Select the six public databases listed above. All off selects AlphaFold DB 50 and PDB.
alignment_type	enum	3di_aa	3di_aa, tmalign, 3di or lolalign. The service owns its database-search parameters.
evalue_threshold	number	optional	Local modes only, from 0 to 10000. Blank uses 10000 for pairwise, 0.001 for monomer clustering, or the native default for other local workflows. A supplied value is passed unchanged. TMalign's result score is not an E-value.
min_seq_id	number	optional	Minimum sequence identity, 0 to 1. Blank uses 0 in pairwise and monomer clustering; otherwise the native default applies.
tmscore_threshold	number	optional	Minimum TM-score, 0 to 1, with normalization selected below. The same blank-value rule applies.
lddt_threshold	number	optional	Minimum LDDT, 0 to 1. Blank uses 0 in pairwise and monomer clustering. Interface workflows retain their native 0.2 default.
coverage	number	optional	Covered-residue fraction, 0 to 1. Blank uses 0 in pairwise and monomer clustering. An entered zero is explicitly forwarded, including during clustering.
remote_iterative	boolean	false	Requests iterative database search from the public service.
remote_taxfilter	string	optional	Native taxonomy-ID expression, such as 9606 or !9606, for databases with taxonomy.
output_format	enum	tsv	Local search/comparison output: tsv, html, sam or pdb. PDB output contains native superposed C-alpha coordinates. Non-TSV runs show their file list in the Data tab.
format_output	string	optional	Comma-separated native TSV columns. Optional fields include qaln, taln, cigar, lddtfull, rmsd, u, t, qcov and tcov. Complex fields require complex or interface search.

Native scientific settings

These optional numeric entries are passed to the corresponding native option. Blank preserves the source default. Controls appear only for workflows that accept them.

Parameter	Type	Default	Description
sensitivity	number	optional	Native -s, from 1 to 9.5. Higher values increase prefilter sensitivity.
max_seqs	integer	optional	Maximum candidates per query passed by the prefilter, at least 1. This can affect sensitivity.
num_iterations	integer	optional	Native search iterations; 0 selects native automatic behavior.
cov_mode	integer	optional	0: query and target; 1: target; 2: query; 3: target/query length; 4: query/target length; 5: shorter/longer length.
seq_id_mode	integer	optional	Identity normalization: 0 alignment length, 1 shorter sequence, 2 longer sequence.
tmscore_threshold_mode	integer	optional	TM threshold normalization: 0 alignment, 1 query length, 2 target length.
tmalign_hit_order	integer	optional	TM-score ordering: 0 mean, 1 query, 2 target, 3 minimum, 4 maximum of query/target scores.
prefilter_mode	integer	optional	0 k-mer/ungapped, 1 ungapped, 2 no filter, 3 ungapped and gapped.
exact_tmscore	integer	optional	0 approximate or 1 exact native TM-score calculation.
tmalign_fast	integer	optional	0 or 1 for native fast TMalign.
exhaustive_search	integer	optional	1 bypasses the prefilter for exhaustive comparison; 0 uses the native prefilter.
mask_bfactor_threshold	number	optional	Mask residues for seeding below this B-factor threshold, from 0 to 100.
mask, mask_lower_case	integer	optional	0 or 1 for native low-complexity and lowercase masking.
min_aln_len	integer	optional	Minimum alignment length, at least 0.
alt_ali	integer	optional	Number of alternative alignments, at least 0.
cluster_mode	integer	optional	Native algorithm: 0 set cover, 1 connected components, 2 or 3 greedy sequence-length clustering.
cluster_search	integer	optional	1 expands native clustered targets after representative search; 0 disables expansion. Plain uploaded collections have no precomputed cluster-member mapping.
multimer_tm_threshold, chain_tm_threshold, interface_lddt_threshold	number	optional	Native complex, individual-chain and interface similarity thresholds, from 0 to 1.
interface_distance	number	optional	Interface C-alpha distance cutoff in angstroms.
interface_min_residues	integer	optional	Minimum residues per chain retained in an interface.

Downloads

All modes retain their original native results and complete diagnostics. Local TSV output retains source column names and precision; the convenience table labels identity as a percentage and identifies TM-score normalization. Complex searches also retain their native assignment report and transforms. Clustering retains assignments and every native sequence output. HTML, SAM and superposed PDB choices use the native output writer directly.

The Data tab previews up to 500 local rows within the result-metadata size budget. A notice identifies incomplete previews. The complete local TSV and native scientific data remain downloadable, including long alignments that cannot fit in a preview. This presentation limit does not filter the native search or clustering output.

Remote searches retain each query response, the complete result archive, the query lookup, database catalog, submitted options and service provenance. The remote engine revision is not exposed by the service; the local executable identity does not identify that engine. Older saved jobs retain their original settings and result interpretation.

Understanding the results

FoldSeek returns different metrics depending on the mode. Database search provides probability scores and E-values, while local comparison provides detailed structural metrics.

Database search results

When searching against AlphaFold DB or PDB, results include:

• Probability: Confidence score from 0 to 1 indicating match quality. Higher values represent more confident structural matches. This is distinct from TM-score and is calculated by the FoldSeek search algorithm.
• E-value: Expectation value representing the number of hits with equal or better scores expected by chance. Lower E-values indicate more significant matches. Values below 0.001 are typically considered confident hits.
• Identity %: Percentage of aligned residues with identical amino acids. This shows sequence conservation in addition to structural similarity.
• Alignment length: Number of residues aligned between query and target structures.

TMalign instead displays a TM-score, with higher values indicating greater structural similarity. Its native score column contains the query TM-score multiplied by 100 and truncated to an integer, not a bit score. Database TMalign results are ordered by their TM-score; other database results use the maintained descending probability order.

Local comparison results

When comparing structures locally (pairwise or clustering mode), FoldSeek calculates:

TM-score (Template Modeling score) measures global structural similarity on a scale of 0 to 1:

TM-score	Interpretation
< 0.17	Random, unrelated structures
0.17 - 0.5	Some structural similarity
> 0.5	Same fold
1.0	Identical structures

A TM-score of 0.5 is a commonly used guide for shared fold similarity. Interpretation depends on normalization and protein length. Alignment-, query- and target-normalized scores are different quantities. Native alignment-normalized values can occasionally exceed 1 because of the source's alignment-length convention; ProteinIQ preserves them without clamping.

LDDT (Local Distance Difference Test) evaluates local structural accuracy without requiring superposition. It compares interatomic distances rather than absolute positions, making it robust to domain movements.

LDDT	Interpretation
> 0.9	Excellent local agreement
0.7 - 0.9	Good local structure
0.5 - 0.7	Moderate agreement
< 0.5	Poor local similarity

LDDT is particularly useful for multi-domain proteins where global superposition may be misleading.

Sequence identity: The fraction of aligned positions with identical amino acids. High sequence identity with low structural similarity may indicate conformational changes. Low sequence identity with high TM-score indicates structural conservation despite sequence divergence.

Use cases

Database search

• Functional annotation: Find proteins with similar folds to infer function
• Evolutionary analysis: Discover distant homologs undetectable by sequence
• Template identification: Find templates for homology modeling
• Novel fold detection: Check if your structure represents a new fold

Local comparison & clustering

• Fold classification: Group structures into families based on 3D similarity
• Redundancy removal: Create non-redundant structure datasets for training ML models
• Quality assessment: Compare predicted structures to known templates
• Conformational analysis: Identify structural changes between states

Limitations

FoldSeek excels at finding structural similarity but has some constraints:

• Requires atomic coordinates (PDB or mmCIF format)
• 3Di encoding may miss some similarities in highly flexible regions
• Clustering is greedy and results depend on representative selection
• Very short structures (< 30 residues) may produce unreliable scores
• Database search depends on the FoldSeek web server and may take 1-5 minutes