FreeBindCraft

FreeBindCraft webserver overview

FreeBindCraft designs protein binders for a selected target structure. This separate tool uses the independently maintained FreeBindCraft fork without PyRosetta. It uses AlphaFold2 and ProteinMPNN for design, OpenMM relaxation and FASPR side-chain packing, then ranks accepted designs by i_pTM or ipSAE.

ProteinIQ also offers BindCraft, which uses restrained minimization without side-chain repacking. Both hosted tools omit Rosetta/PyRosetta due to licensing. Projects that require Rosetta relaxation, energies, or filtering should contact sales to discuss licensing and requirements. Neither hosted method establishes equivalence to the published Rosetta-inclusive pipeline.

See the BindCraft online guide, including FreeBindCraft alternatives for the workflow, a PD-L1 input exercise, and interpretation.

Pricing

The standard quote reserves 5,400 credits for two hours on an A100 80 GB GPU. The rate is 45 credits per minute; actual billable runtime is settled after the run and unused reserved credits are returned. The exact quote appears before submission. Full diagnostic artifacts do not change the standard compute reservation. Design count, target size, binder length, and iteration settings can change runtime but do not change the standard compute reservation.

Inputs

Input	Accepted value
Target protein	One .pdb or .ent structure, up to 50 MB, or a PDB fetched from RCSB.
Target chains	One chain ID or a comma-separated list such as A,B.
Hotspot residues	Optional residue numbers, ranges, or chain-qualified selectors such as 56, A15, or A1-10.

Settings

The tables show defaults for the default design protocol. Choosing another protocol or advanced preset updates the affected controls to that preset's values, including peptide lengths of 8 to 25 residues. Your subsequent edits are used exactly as entered, including false, zero and values that match the default protocol.

Leave quality thresholds blank to use the selected filter preset. Entering a threshold replaces the corresponding native filters. With preset overrides enabled, custom advanced and filter JSON takes precedence over individual controls; protected execution settings cannot be overridden. advanced.json and filters.json record the effective settings.

Core settings

Parameter	Type	Default	Description
design_protocol	enum	default	Default is recommended for miniprotein binders. Beta-sheet promotes beta-sheeted proteins and usually needs more sampling. Peptide uses the source peptide presets and peptide filters. Choices: default, beta_sheet, peptide.
rank_by	enum	i_pTM	FreeBindCraft ranks accepted designs by i_pTM by default. ipSAE is an optional AlphaFold interface confidence measure. Choices: i_pTM, ipSAE.
num_designs	integer	1	Number of final binder designs to generate. More designs = longer runtime. Range: 1 to 100.
length_min	integer	40	Minimum number of residues in the designed binder. Range: 8 to 100. Peptide default: 8.
length_max	integer	80	Maximum number of residues in the designed binder. Range: 8 to 300. Peptide default: 25. Must be at least the minimum binder length.
target_chains	string	A	Chain ID(s) from the target PDB to design binders against. Use comma for multiple chains (e.g., A,B).
hotspot_residues	string	optional	Specific residues, ranges, or whole chains to focus binding. Unchained numbers use the first selected target chain. Leave empty for automatic selection.

Protocol and model controls

Parameter	Type	Default	Description
prediction_protocol	enum	default	Default performs single-sequence binder prediction. Hard target uses initial guess to improve difficult complex prediction, with possible bias. Choices: default, hardtarget.
interface_protocol	enum	alphafold2	AlphaFold2 generates the interface by default. MPNN uses soluble ProteinMPNN to optimize the interface. Choices: alphafold2, mpnn.
template_protocol	enum	default	Default allows limited target flexibility. Flexible masks template sequence for more backbone and sidechain flexibility. Choices: default, flexible.
filter_type	enum	auto	Filter preset for design acceptance. Auto uses peptide filters for peptide design and default filters otherwise. Choices: auto, default_filters, relaxed_filters, peptide_filters, peptide_relaxed_filters, no_filters.
design_algorithm	enum	auto	Select the preset algorithm with auto or explicitly choose an optimization schedule. Choices: auto, 4stage, 3stage, 2stage, greedy, mcmc.
use_multimer	boolean	true	Use AlphaFold2-multimer for better interface prediction. Recommended for most cases.
sample_models	boolean	true	Randomly sample AlphaFold2 parameter sets during design, as recommended by FreeBindCraft to reduce overfitting.

Quality thresholds

Parameter	Type	Default	Description
plddt_threshold	number	preset	Minimum accepted pLDDT shown on the familiar 0-100 scale. FreeBindCraft applies this internally as a normalized 0-1 filter threshold to the Average/1/2 pLDDT checks. Range: 50 to 95.
ptm_threshold	number	preset	Minimum accepted pTM threshold. FreeBindCraft uses normalized 0-1 values internally and applies this to the Average/1/2 pTM filters. Range: 0.3 to 0.9.
i_ptm_threshold	number	preset	Minimum accepted interface pTM threshold. FreeBindCraft uses normalized 0-1 values internally and applies this to the Average/1/2 i_pTM filters. Range: 0.3 to 0.9.
i_pae_threshold	number	preset	Maximum accepted interface pAE threshold for Average/1/2 i_pAE filters. Lower is stricter. Range: 0.1 to 0.8.
binder_rmsd_threshold	number	preset	Maximum binder-alone RMSD threshold for Average/1-5 Binder_RMSD filters. Lower is stricter. Range: 1 to 8.
include_unaccepted	boolean	true	With full diagnostic artifacts enabled, also display evaluated designs that did not pass the configured FreeBindCraft filters. These are exploratory candidates, not accepted FreeBindCraft designs.

Preset overrides

Parameter	Type	Default	Description
preset_options	boolean	false	Override the FreeBindCraft advanced and filter JSON presets directly.
advanced_preset	enum	default_4stage_multimer	FreeBindCraft advanced JSON preset. Leave preset overrides off to let binder mode choose the matching preset automatically. Choices: default_4stage_multimer, default_4stage_multimer_flexible, default_4stage_multimer_hardtarget, default_4stage_multimer_flexible_hardtarget, default_4stage_multimer_mpnn, default_4stage_multimer_mpnn_flexible, default_4stage_multimer_mpnn_hardtarget, default_4stage_multimer_mpnn_flexible_hardtarget, betasheet_4stage_multimer, betasheet_4stage_multimer_flexible, betasheet_4stage_multimer_hardtarget, betasheet_4stage_multimer_flexible_hardtarget, betasheet_4stage_multimer_mpnn, betasheet_4stage_multimer_mpnn_flexible, betasheet_4stage_multimer_mpnn_hardtarget, betasheet_4stage_multimer_mpnn_flexible_hardtarget, peptide_3stage_multimer, peptide_3stage_multimer_flexible, peptide_3stage_multimer_mpnn, peptide_3stage_multimer_mpnn_flexible.
filter_preset	enum	default_filters	FreeBindCraft filter JSON preset. Leave preset overrides off to let binder mode choose the matching preset automatically. Choices: default_filters, relaxed_filters, peptide_filters, peptide_relaxed_filters, no_filters.
custom_advanced_json	string	optional	Custom advanced JSON object; takes precedence over individual controls. Execution paths and protected controls cannot be overridden.
custom_filters_json	string	optional	Custom filters may use computed FreeBindCraft metrics, including Average_ipSAE and 1_ipSAE through 5_ipSAE. Rosetta energy, packing, and hydrogen-bond values are placeholders and cannot be used as measured filter criteria.

Runtime controls

Parameter	Type	Default	Description
gpu_tier	enum	a100_80gb	FreeBindCraft runs on A100 80 GB for this deployment. Choices: a100_80gb.
max_trajectories	integer	0	Maximum hallucination trajectories before stopping. Use 0 for default behavior with only the runtime cap. Range: 0 to 10000.
enable_rejection_check	boolean	true	Stop early if the accepted-design rate stays below the configured acceptance rate after monitoring starts.
acceptance_rate	number	0.01	Minimum fraction of trajectories expected to yield accepted designs once rejection monitoring starts. Range: 0 to 0.5.
start_monitoring	integer	600	Trajectory count before acceptance-rate monitoring can stop the run. Range: 1 to 5000.

Template options

Parameter	Type	Default	Description
predict_initial_guess	boolean	false	Bias prediction with binder atom positions; use cautiously when redesigns fail validation.
predict_bigbang	boolean	false	Bias atom initialization for large target/design systems. Useful for larger complexes.
rm_template_seq_design	boolean	false	Remove target template sequence during binder design.
rm_template_seq_predict	boolean	false	Remove target template sequence during validation prediction.
rm_template_sc_design	boolean	false	Remove target sidechains from the design template.
rm_template_sc_predict	boolean	false	Remove target sidechains from the validation prediction template.

ProteinMPNN options

Parameter	Type	Default	Description
mpnn_fix_interface	boolean	true	Keep the interface generated by the trajectory fixed during ProteinMPNN redesign.
num_seqs	integer	20	Number of ProteinMPNN sequences sampled per accepted trajectory. Higher values cost more. Range: 1 to 200.
max_mpnn_sequences	integer	2	Maximum passing MPNN redesigns saved for each trajectory. Higher values can produce more outputs and more validation work. Range: 1 to 10.
sampling_temp	number	0.1	ProteinMPNN sampling temperature. Higher values sample more diverse sequences. Range: 0 to 1.
backbone_noise	number	0	Backbone noise used during ProteinMPNN sampling. Range: 0 to 0.05.
mpnn_weights	enum	soluble	ProteinMPNN weights used by FreeBindCraft. Choices: soluble, original.
omit_AAs	string	C	Comma-separated amino acids to avoid; cysteine is omitted by default.
force_reject_AA	boolean	false	Force rejection when a design contains any amino acids listed in omit_AAs.

Design bias and loss weights

Parameter	Type	Default	Description
weights_helicity	number	-0.3	Helicity loss weight; negative values favor beta sheets and positive values favor helices. Range: -3 to 1.
random_helicity	boolean	false	Sample a random helicity weight per trajectory instead of using the fixed value.
use_i_ptm_loss	boolean	true	Use interface pTM as an optimization loss.
weights_iptm	number	0.05	Weight for the interface pTM optimization loss. Range: 0 to 1.
use_rg_loss	boolean	true	Use compactness loss for the binder.
weights_rg	number	0.3	Weight for the radius-of-gyration loss. Range: 0 to 2.
use_termini_distance_loss	boolean	false	Try to minimize distance between the binder N- and C-termini. Useful for grafting.
weights_termini_loss	number	0.1	Weight for the N/C-terminus distance loss. Range: 0 to 2.

Iterations and recycles

Parameter	Type	Default	Description
soft_iterations	integer	75	Iterations that optimize continuous amino-acid preferences. Range: 0 to 500.
temporary_iterations	integer	45	Iterations that optimize softened amino-acid probabilities. Range: 0 to 500.
hard_iterations	integer	5	Iterations using discrete amino-acid choices. Range: 0 to 200.
greedy_iterations	integer	15	Iterations that test sequence mutations to reduce the design loss. Range: 0 to 500.
greedy_percentage	number	1	Percentage of binder positions mutated during a greedy step. Range: 0 to 20.
num_recycles_design	integer	1	AlphaFold recycles during design. Higher values cost more. Range: 1 to 12.
num_recycles_validation	integer	3	AlphaFold recycles during validation. Higher values cost more. Range: 1 to 12.
optimise_beta	boolean	true	Apply extra optimization when a beta-sheet trajectory is detected.
optimise_beta_extra_soft	integer	0	Additional soft iterations for beta-sheet optimization. Range: 0 to 300.
optimise_beta_extra_temp	integer	0	Additional temporary iterations for beta-sheet optimization. Range: 0 to 300.
optimise_beta_recycles_design	integer	3	AlphaFold recycles during beta-sheet design optimization. Range: 1 to 12.
optimise_beta_recycles_valid	integer	3	AlphaFold recycles during beta-sheet validation. Range: 1 to 12.

Contact and confidence weights

Parameter	Type	Default	Description
weights_plddt	number	0.1	Weight for local confidence in the designed binder. Range: 0 to 2.
weights_pae_intra	number	0.4	Weight for predicted aligned error within the binder. Range: 0 to 2.
weights_pae_inter	number	0.1	Weight for predicted aligned error between target and binder. Range: 0 to 2.
weights_con_intra	number	1	Weight for contacts within the binder. Range: 0 to 5.
weights_con_inter	number	1	Weight for contacts between target and binder. Range: 0 to 5.
intra_contact_distance	number	14	Within-binder contact distance cutoff in angstroms. Range: 4 to 30.
inter_contact_distance	number	20	Target-binder contact distance cutoff in angstroms. Range: 4 to 40.
intra_contact_number	number	2	Desired contacts per residue within the binder. Range: 1 to 8.
inter_contact_number	number	2	Desired contacts per residue between target and binder. Range: 1 to 8.

Full diagnostic artifacts

Parameter	Type	Default	Description
artifact_options	boolean	false	Return full trajectory animations, plots, rejected PDBs, and optional archive outputs. This can substantially increase the number and size of downloaded files. Leave off for compact accepted-design and CSV diagnostic outputs.
save_design_animations	boolean	true	Save design animations when full diagnostics are enabled.
save_design_trajectory_plots	boolean	true	Save trajectory plots when full diagnostics are enabled.
save_trajectory_pickle	boolean	false	Save trajectory arrays as pickle files; can use substantial storage.
save_mpnn_fasta	boolean	false	Save ProteinMPNN sequences as FASTA files in full diagnostics.
zip_animations	boolean	true	Archive animation files when full diagnostics are enabled.
zip_plots	boolean	true	Archive plot files when full diagnostics are enabled.
remove_unrelaxed_trajectory	boolean	true	Remove unrelaxed trajectory PDBs after scoring.
remove_unrelaxed_complex	boolean	true	Remove unrelaxed complex PDBs after scoring.
remove_binder_monomer	boolean	true	Remove binder-alone PDBs after scoring.

Outputs

Output	Meaning
Ranked complex PDBs	Accepted binder-target structures in the source ranking order.
Design CSV files	final_design_stats.csv, mpnn_design_stats.csv, and trajectory_stats.csv retain the original source columns and values.
Rejection record	failure_csv.csv records filter failures.
Run files	freebindcraft.log and the resolved settings.json, filters.json, and advanced.json remain downloadable. Plots, animations, FASTA files, and diagnostic structures are included when full diagnostic artifacts are enabled.

Reading the results

Average_i_pTM is the default ranking metric; Average_ipSAE can be selected instead. pLDDT, pTM, PAE, interface geometry, SASA, and structure files are available when the source computes them. FreeBindCraft uses exposed-area weighting for surface hydrophobicity, including glycine in its hydrophobic residue set. Its binder RMSD uses C-alpha atoms and may truncate unequal residue lists.

The source writes fixed compatibility values for Rosetta energy, packing, and hydrogen-bond fields. These include dG, binder energy, PackStat, hydrogen-bond counts and percentages, and dG/dSASA. ProteinIQ omits those fields from displayed measurements and custom filter controls, but keeps the original CSV files for audit. The source default filters treat those fields as passing; an accepted design has not passed an independent Rosetta energy or hydrogen-bond assessment.

If shape complementarity, SASA, FASPR packing, or OpenMM relaxation falls back to a substitute value or structure, ProteinIQ marks the run failed and keeps its diagnostic record and any previously written accepted structures and sequences. Partial outputs do not mean the run completed successfully. The fork has not established experimental equivalence to PyRosetta-inclusive BindCraft. Treat computational candidates as hypotheses for further evaluation.