Use DISCO Online
Commercially Available DISCO No-Code Web Server
DISCO: De Novo Multimodal Protein Co-Design for Arbitrary Biomolecules
DISCO (DIffusion for Sequence-structure CO-design) is a cutting-edge multimodal framework that simultaneously designs protein sequences and 3D structures de novo. By optimizing both modalities in a unified generative process, DISCO creates highly diverse, co-designable, and functional proteins tailored to arbitrary molecular targets without requiring sequence inverse-folding.
Key Capabilities of DISCO
Multimodal Sequence-Structure Co-Design: Unlike traditional decoupled pipelines that create a backbone before mapping a sequence, DISCO models sequences and 3D atomic coordinates simultaneously via a single deep neural network, enabling sequence objectives to inform structure formation and vice versa.
Conditioning on Arbitrary Molecular Contexts: DISCO can co-fold and adapt its binding geometry around small molecules, metallocofactors, reactive intermediates, and nucleic acids (DNA/RNA).
Active Site Innovation Without Pre-Specified Motifs: Generate enzymes without pre-specifying fixed Arrangements of catalytic residues or theozymes. DISCO explores entirely novel active-site geometries previously unsampled by natural evolution.
Inference-Time Property Steering: Utilizing a novel multimodal Feynman-Kac Corrector (FKC) framework, DISCO can scale and tilt sampling distributions to optimize specific biophysical traits, such as enriched disulfide bond counts, fine-grained contacts, or exclusive target-binding specificity.
Experimentally Validated Performance
DISCO has been rigorously tested both in silico and in vitro, surpassing typical starting points for directed evolution and expanding genetically encodable chemistry:
90% Monomer Refolding Success: In unconditional monomer de novo design, approximately 90% of generated sequences successfully fold back to within 2 Å RMSD of their predicted structures.
High-Yield New-to-Nature Catalysis: DISCO-designed heme enzymes successfully catalyze four separate, new-to-nature carbene-transfer transformations (alkene cyclopropanation, spirocyclopropanation, B-H insertion, and C(sp^3)-H alkylation) with yields as high as 98% and total turnover numbers (TTN) up to 5,170.
Evolvable Baselines: Wet-lab random mutagenesis confirmed that DISCO’s structurally novel enzymes occupy accessible fitness landscapes, allowing immediate activity improvements through a single round of directed evolution.
What is Tamarind Bio?
Tamarind Bio is a leading computational biology platform designed to accelerate therapeutic research, enzyme engineering, and molecular innovation. By hosting state-of-the-art generative deep learning algorithms like DISCO in an accessible ecosystem, Tamarind Bio removes the infrastructure bottlenecks of machine-learning-driven protein design. Whether you are scaffolding functional sites, creating specific DNA binders, or pioneering biocatalysts for green chemistry, Tamarind Bio provides the compute scaling, optimization tools, and predictive models required to bring computational designs to experimental reality
How to Use DISCO on Tamarind Bio
Running your protein co-design campaign with DISCO on Tamarind Bio is simple, programmatic, and completely code-free:
Define Your Input Modalities: Provide the coordinates and bonding properties of your target biomolecule or reaction intermediate (such as Molecular SMILES, metal cofactors, or DNA/RNA sequences).
Specify Target Parameters: Enter your desired protein chain length and configure any sequence-structure steering objectives (such as optimizing specific residue contacts or applying specificity guidance against decoy molecules).
Generate and Co-Fold: DISCO will simultaneously unmask candidate protein sequences while dynamically denoising and adjusting the coordinates of both the protein and your conditioning ligands to capture induced conformational changes.
Review Filtered Outputs: Receive candidate sequence-structure pairs pre-ranked by robust structural confidence measures (including predicted TM-scores (pTM) and pairwise aligned error (PAE)), fully optimized for immediate wet-lab expression or directed evolution.