Use SwitchCraft Online
Commercially Available SwitchCraft No-Code Web Server
SwitchCraft: Programmatic Design of State-Switching and Multistate Proteins
SwitchCraft is a programmatic framework for designing state-switching proteins, meaning proteins that activate, deactivate or switch between functional states depending on a ligand effector. It was introduced by Bowen Jing, Mihir Bafna and colleagues at MIT and UT Austin, in a paper presented at ICML 2026. Code is available on GitHub.
Protein language models and structure-based diffusion models have transformed protein design, but natural function is often multistate: motors, rotary ATP synthase and allosteric enzymes all change conformation. Language models give only coarse control, while structure generators produce individual static structures. No large dataset pairs sequences with detailed multistate behaviour, so purely data-driven approaches are not available.
SwitchCraft instead backpropagates through a structure prediction model (Boltz-1) using compositional loss functions, so a designer can specify several states and what should differ between them.
Property | Detail |
|---|---|
Method | Gradient-based sequence optimisation through Boltz-1 with compositional design losses |
Specification | States, each with a folding context (fixed molecules), loss functions, a design mask and optional motif sequence |
Effectors | Small molecules, metal ions and DNA |
Validation | In silico, by Boltz-1 predictions in each state |
Code | github.com/bjing2016/switchcraft |
Multistate Protein Design Tasks Supported by SwitchCraft
Positive and negative allostery: a motif is disrupted without a ligand and scaffolded with it, or the reverse.
Motif switching: two motifs toggle upon ligand binding.
Induced binding: a designed protein binds a partner only in the presence of an effector such as Ca2+.
Ligand modification: a large conformational shift when a bound ligand changes, for example heme to oxygenated heme.
Ligand discrimination: three distinct conformations depending on which of two ligands is bound.
SwitchCraft In Silico Results and Biosensor Design Workflow
Experiment | Reported result |
|---|---|
Positive/negative allostery | 11 of 24 RFdiffusion benchmark motifs with at least one success, 100 designs per motif, ligand and specification |
Motif switching (OQO effector) | 3 of 100 designs scaffold both motifs correctly |
Heme and oxygen switch | 10 of 558 designs show the intended change; example 3.8 Å shift |
Ligand discrimination (OQO, Ca2+) | 12 of 465 designs successful |
Biosensor conformation switchers (SAM, cGMP, ATP) | 89 of 13,858 designs pass stringent criteria |
De Novo Fluorescent Biosensor Workflow
Biosensors typically fuse circularly permuted GFP (cpGFP) to a ligand-responsive switch. SwitchCraft designs conformation switchers using contact, binding and conformational-change losses, then inserts cpGFP at sites with large backbone changes and co-folds the constructs. Screening for changes in chromophore contacts left 44 designs, including a SAM sensor in which Glu74 moves away from the chromophore on ligand binding.
What is Tamarind Bio?
Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.
The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:
A user-friendly graphical interface for setting up and launching experiments
A robust API for integration into existing research pipelines
An automated system for managing and scaling computational resources
Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.
Accelerating Discovery with SwitchCraft on Tamarind Bio
Fluorescent biosensors: design switches for small-molecule analytes such as SAM, cGMP and ATP.
Allosteric regulation: design proteins whose motifs turn on or off with a ligand.
Ligand-gated binding: design binders that engage a partner only with an effector present.
Multistep function: explore proteins with several ligand-dependent states, as in enzyme cycles.
How to Use SwitchCraft on Tamarind Bio
Open the tool: log in to tamarind.bio and select SwitchCraft.
Define the states: specify the folding context for each state, such as apo and ligand-bound, if exposed.
Specify the objective: choose the design behaviour, for example allostery, induced binding or a conformational change, if exposed.
Set design parameters: protein length, designed positions and any fixed motif, if exposed.
Run many trajectories: success rates are low, so generate many designs.
Download and filter: filter by effector ipTM, pLDDT, within-state and cross-state RMSD.
Use downstream: prioritise designs for experimental testing.
Things to Keep in Mind
Results are in silico; the authors list experimental validation as future work.
Success rates remain low in absolute terms, so large numbers of designs are needed.
Evaluation relies on Boltz-1 predictions, which rarely predict multistate behaviour for natural proteins, so filtering is important.
Some designs show large conformational changes that appear implausible without ligand unbinding.
Source: Jing B, Bafna M, Parsan A, Ni HM, Kwabi-Addo D, Bryson B, Klivans A, Berger B. SwitchCraft: A Programmatic Framework for Designing State-Switching Proteins. ICML 2026 (PMLR 306); arXiv:2605.31236. Code: https://github.com/bjing2016/switchcraft.