BioEmu
v1.4.1Sample independent protein equilibrium conformations.
BioEmu2 is available
BioEmu 1.4.1 samples equilibrium conformations of a single protein monomer. ProteinIQ accepts a protein sequence, whose alignment BioEmu retrieves from ColabFold, or a supplied A3M alignment, and returns native structures, ensemble coordinates, raw sample batches and reproducibility files. Independent samples do not represent motion over time or a time-resolved molecular dynamics trajectory.
New jobs are currently disabled while validation is completed. The reference below describes the implemented controls and current pricing estimates. Side-chain reconstruction and MD relaxation are outside this tool's scope.
Runs cost 57 credits per minute of measured runtime. The minimum reservation is 57 credits, not a minimum final charge. Completed runs are charged in proportion to elapsed time, rounded up to a whole credit; unused reserved credits are returned. Set a spending limit before submission. Each batch job is metered separately.
| Input | Accepted format | Upload limit | Requirements |
|---|---|---|---|
| Protein sequence | One sequence or one-record FASTA, pasted, uploaded or fetched from RCSB | 10 MiB per file | Use this or an alignment. BioEmu retrieves the MSA from the public ColabFold server. |
| Protein sequence and alignment | .a3m, upload or pasted text | 50 MiB per file | Instead of a sequence. The first record is the query monomer; the rest are its alignment. |
| Native denoiser configuration | .yaml, .yml, .json, upload or pasted YAML | 1 MiB per file | Optional. Overrides the selected denoiser's configuration. |
| Steering reference PDB | .pdb | 50 MiB per file | Optional. Required when the configuration refers to reference.pdb. |
| Previous native samples | .zip | 50 MiB; 1 GiB once expanded | Optional. Root-level sequence.fasta and batch_<start>_<end>.npz files only. |
Account upload limits may be lower. Each job accepts one file per input role. No tool-specific maximum sequence length or sample count is configured; larger requests may exhaust available memory or runtime.
The query must contain the 20 uppercase standard amino acids. Gaps, ambiguous residues and multimer separators are not accepted in the query. A3M insertions in alignment rows are preserved. With a sequence, BioEmu retrieves its alignment from ColabFold; with an A3M, it uses yours.
Continuation must use the same query sequence as the saved native state. Existing batches remain unchanged, and Number of samples specifies the total target, including those batches. The continuation.zip download provides the required files.
| Parameter | Type | Default | Description |
|---|---|---|---|
Number of samples (num_samples) | integer | Required | At least 1; no preset value. Physical filtering can leave fewer conformations. |
Checkpoint (model_name) | enum | bioemu-v1.1 | bioemu-v1.0 (preprint), bioemu-v1.1 (Science paper), or bioemu-v1.2 (extended training). Custom checkpoints are unavailable. |
Denoiser (denoiser_type) | enum | dpm | DPM uses 50 steps; heun uses 100. An uploaded configuration takes precedence. |
Filter unphysical samples (filter_samples) | boolean | true | Applies BioEmu's clash and chain-continuity filters; raw batches are retained. |
Base random seed (base_seed) | string | Optional | Blank uses BioEmu's clock seed. Integer text from -9223372036854775808 to 18446744073709551615 preserves full precision. |
| Job name | string | Optional | A label for the run. |
Uploaded configuration supports BioEmu's DPM, Heun, SMC and FKC solvers, including native step, noise and steering controls. It must select an allowed BioEmu solver through _target_ and set _partial_: true; native boolean spellings such as yes are also accepted.
Reference-based controls must use reference_pdb: reference.pdb with the reference input. If an output_dir is specified, it must be diagnostics. Arbitrary Python targets, external file paths and configuration interpolation are not accepted. Configuration bytes are retained in the downloads.
Conformations displays the topology with its ensemble coordinates. Files provides individual downloads, including native and reproducibility artifacts.
| Download | Contents |
|---|---|
native/topology.pdb | Native structure topology used with the ensemble coordinates. |
native/samples.xtc | Ensemble coordinates, paired with the topology PDB. |
native/samples_unphysical.xtc | Returned instead of samples.xtc when no conformations pass the physical filter, with a warning. |
native/batch_*.npz | All unfiltered positions, orientations and query sequence in native batch order. |
native/sequence.fasta | The sampled query sequence. |
records/continuation.zip | Unchanged native FASTA and raw batches for continuation. |
records/denoiser.yaml | The supplied configuration or selected native default configuration. |
records/run.log | Full sampling log, diagnostics and native seed information. |
records/provenance.json | Settings, command, input hashes, source identity, model identities and dependency records. |
| Input and attribution files | Submitted inputs, BioEmu license, model card and third-party notices, including AlphaFold2 attribution. |
Handled sampling failures retain available native files, logs and provenance. A continuation archive is included when FASTA and batch files are available; the job remains failed and does not claim a complete ensemble.
Ensemble order is not chronological and has no kinetic interpretation. The physical filter checks geometry; passing it does not establish experimental accuracy. Raw NPZ batches include rejected samples, so their count can exceed the filtered XTC count. samples_unphysical.xtc contains samples that failed those checks and should be interpreted with the accompanying warning and log.