Use Boltz-2 (AlphaFold3)

Official Neurosnap webserver for accessing Boltz-2 (AlphaFold3) online.

Protein Folding Molecular Docking

Overview

Boltz-2 is the next-generation open-source deep-learning model family that reaches AlphaFold3-level accuracy for 3-D biomolecular complex prediction and adds a dedicated head for near-FEP binding-affinity estimation. Architectural depth, physics-aware Boltz-steering potentials, ensemble-based training, and fine-grained controllability together deliver fast, clash-free structures and quantitative interaction scores across proteins, nucleic acids, and small molecules. Released under the permissive MIT license, Boltz-2 ships with full training code, pretrained weights, and curated datasets to accelerate discovery in drug design and structural biology.

Neurosnap Overview

The Boltz-2 (AlphaFold3) online webserver allows anybody with a Neurosnap account to run and access Boltz-2 (AlphaFold3), no downloads required. Information submitted through this webserver is kept confidential and never sold to third parties as detailed by our strong terms of service and privacy policy.

View Paper

Features

  • Dedicated affinity-prediction head with near-FEP accuracy for binding-energy estimation.
  • Ensemble-based training on NMR and MD trajectories to capture conformational dynamics.
  • Fine-grained controllability via experimental-method, template, and pocket/distance constraints.
  • Boltz-steering physics potentials yield clash-free, stereochemically correct structures.
  • Deeper mixed-precision PairFormer (64 layers) with larger crops for faster inference on large complexes.
  • Extended input representation (cyclic/modified flags, bond-order, symmetric positional encodings).
  • Millions of curated biochemical-assay measurements for reliable affinity and generalisation.
  • Random MSA dropout and ensemble supervision boost single-sequence robustness.

Statistics

Neurosnap periodically calculates runtime statistics based on job execution data. These estimates provide a general guideline for how long your job may take, but actual runtimes can vary significantly depending on factors like input size or settings used.

Statistic Value
Credit Usage Rate loading...
Estimated Total Cost loading...
Runtime Mean loading...
Runtime Median loading...
Runtime Standard Deviation loading...
Runtime 90th Percentile loading...
Runtime Longest loading...

API Request

Access Boltz-2 (AlphaFold3) using the Neurosnap API by sending a request using any programming language with HTTP support. To safely generate an API key, visit the API tab of your overview page.

Job Note

Provide a name or description for your job to help you organize and track its results. This input is solely for organizational purposes and does not impact the outcome of the job.

Configuration & Options

Service Inputs

The amino acid, DNA, and RNA sequences corresponding to the molecules you want to predict. For complex prediction simply provide multiple sequences in this input and they will all be folded at the same time together. For DNA and RNA sequences enter them in the 5′-->3′ direction. For DNA do not include the complementary sequence as they are automatically generated.

Input small molecules to include in prediction. All inputs are converted to SMILES format, with SMILES strings being the preferred input for optimal results.

This is an optional input for specifying any biopolymer chains (must be a protein or nucleotide and a not ligand) that are cyclic. Values must be sequence names defined within the "Input Sequences" input and must be comma delimited for multiple entries.

Residue Modifications Instructions

Specify residue modifications for amino acid, DNA, or RNA sequences. Each row should define one modification, with values separated by colons (:). The format is:

SEQ_ID : RESIDUE_INDEX : CCD_CODE
  • SEQ_ID → The name of the sequence as defined in Input Sequences.
  • RESIDUE_INDEX → The residue position to modify. Use 1 for the first residue.
  • CCD_CODE → The Chemical Component Dictionary (CCD) code of the modification.


🔹 Example Input:

Protein_1:102:MLY
DNA_1:1:5MC
RNA_1:26:PSU


⚠️ Notes:

See above instructions for information on how to use this input.

Pocket Restraints Instructions (Optional)

The Binder Sequence corresponds to the binder chain, while "Pocket Restraints" specifies residues interacting with it.

Specify inter-chain pocket restraints to guide Boltz-1 in folding complexes. These restraints define interactions between a binder sequence and residues in other sequences, influencing the folding process.

Each row should define one pocket restraint, with values separated by colons (:). The format is:

CONTACT_SEQ:CONTACT_RES
  • CONTACT_SEQ → The sequence containing the interacting residue.
  • CONTACT_RES → The position of the residue on CONTACT_SEQ.


🔹 Example Input:

Protein_2:66
Protein_2:78
DNA_seq1:13


⚠️ Notes:

  • Sequence names match those in Input Sequences.
  • Residue numbering starts at 1.
  • The model currently only supports a single binder sequence per pocket restraint, but multiple contact residues can be specified across different sequences.
  • The sequence name of the binder should only be specified if pocket restraints are being used.

Specify the sequence acting as the binder. See above instructions for more details.

Specify residues interacting with the binder sequence. See above instructions for more details.

Covalent Restraints Instructions (Optional)

Specify covalent bonds between atoms to guide Boltz-1 in complex folding. These restraints define fixed interactions between atoms in different sequences, ensuring structural constraints are maintained.

Each row should define one covalent restraint, with values separated by colons (:). The format is:

SEQ_ID1:RES_ID1:ATOM_NAME1:SEQ_ID2:RES_ID2:ATOM_NAME2
  • SEQ_ID1 → The sequence containing the first atom.
  • RES_ID1 → Residue index on SEQ_ID1.
  • ATOM_NAME1 → Atom name in RES_ID1.
  • SEQ_ID2 → The sequence containing the second atom.
  • RES_ID2 → Residue index on SEQ_ID2.
  • ATOM_NAME2 → Atom name in RES_ID2.


🔹 Example Input:

Protein_1:6:CA:Protein_2:26:CB
Ligand_1:1:N1:Protein_3:45:OG


⚠️ Notes:

  • Sequence names match those in Input Sequences.
  • Residue numbering starts at 1.
  • Atom names must match standardized PDB/CIF naming conventions.
  • Only canonical residues and CCD ligands are supported.
  • Covalent restraints ensure atoms remain bonded during folding but do not enforce bond angles or torsions.

Specify covalent bonds between atoms. See above instructions for more details.

Advanced Settings

Choose the MSA database you want to use. Selecting "single_sequence" will result in the MSA containing only your target sequence. Note if you upload a custom MSA then your selection here will be ignored. In most cases mmseqs2_uniref_env tends to produce the best results. Warning: We query a 3rd party service to perform sequence searches, if you do not wish to share your sequence set this to "single_sequence".

Weather to run the original Boltz-2 model using inference time potentials.

Whether to add the Molecular Weight correction to the affinity value head.

Increased recycling steps tend to produce more accurate predictions but will also greatly increases prediction time. For smaller proteins and monomers we recommend 5 recycles, for bigger proteins we recommend 10 recycles.

The number of sampling steps to use for prediction.

The number of diffusion samples to use for prediction. This controls the number of output structures returned from the model.

The number of sampling steps to use for affinity prediction.

The number of diffusion samples to use for affinity prediction.

The step scale / size is related to the temperature at which the diffusion process samples the distribution. The lower the higher the diversity among samples (recommended between 1 and 2).

Ready to submit your job?

Once you're done just hit the submit button below and let us do the rest.

To submit a job please login or register an account.