Protein & complex prediction
Run single-chain, multi-chain, and ligand-bound predictions through a guided notebook workflow.
AI biomolecular structure prediction
From sequence to structure, affinity, and screening — in one polished notebook workflow.
Boltz2 Notebook packages Boltz2 into a guided, research-ready interface for protein structure prediction, protein–ligand modeling, DNA/RNA-enabled complexes, confidence analysis, and batch-scale runs without local GPU setup.
Python 3.10
CUDA-enabled
Google Colab
MIT License
Production-ready workflow
No local install
# Example job definition
protein_sequence: MSEQNNTEMT...
ligand_smiles: CCO
binder: B
msa_mode: server
template_file: template.pdb
Core outputs
Structures + confidence
Affinity support
Prediction + ranking
Typical runtime
2–10 min on T4
Best for
Research & screening
3
Launch paths
V1 stable, V2 beta, and Batch.
4
Batch input modes
CSV, FASTA, YAML ZIP, and YAML folder workflows.
7-step
Batch run system
From workspace bootstrap to archival export.
Zero-install
Colab-first delivery
Designed to remove local GPU and CLI friction.
Why this notebook exists
Research-grade modeling without the usual setup overhead.
The repository combines notebook UX, Python utilities, release tracking, and batch orchestration into one coherent entry point for structure prediction and interaction analysis.
Affinity-aware analysis
Surface confidence metrics, predicted aligned error, and affinity-oriented outputs in the same workflow.
Advanced multi-entity support
V2 expands into DNA, RNA, PTMs, covalent chemistry, contact conditioning, and template-guided jobs.
Batch screening
Queue many jobs from structured inputs and rank results for large-scale exploration and prioritization.
Interactive outputs
Designed around visualization, analysis summaries, exportable artifacts, and reproducible notebook usage.
Built for accessibility
Use free Colab GPU resources instead of managing a local CUDA installation and runtime configuration.
End-to-end flow
A clear path from biomolecular input to exported results.
01
Setup
Initialize the notebook environment, install dependencies, and configure workspace directories.
02
Build inputs
Provide protein, ligand, DNA/RNA, MSA, and template details through structured parameters.
03
Run Boltz2
Generate MSAs, launch predictions, recycle structures, and produce PDB or CIF outputs.
04
Analyze
Inspect pLDDT, PAE, confidence, and affinity-linked metrics to assess output quality.
05
Visualize
Review structures interactively and compare candidate predictions for downstream interpretation.
06
Export
Download zipped artifacts or save runs to Drive for sharing, reuse, and traceability.
Real-world applications
How teams use Boltz2 Notebook.
Drug Discovery
Predict protein-ligand binding poses and affinity scores to screen compounds without wet lab costs.
Enzyme Engineering
Design multi-chain enzyme complexes with template guidance and constraint conditioning for better function.
Structure Validation
Rapidly validate experimental structures against predictions and compare confidence metrics (pLDDT, PAE).
Notebook lineup
Choose the right surface for the job.
Main entry (Stable)
Recommended
Main Notebook (V1.0.0)
General-use interface for structure prediction, ligand binding, affinity analysis, confidence inspection, and export.
- Multi-chain proteins
- Protein–ligand support
- Affinity analysis and visualization
Beta
Advanced
V2.0.0
Extended modeling stack for more complex jobs, including constraints, templates, DNA/RNA, and PTMs.
- DNA/RNA support
- Custom MSA & templates
- Covalent and contact constraints
Experimental
Batch
Batch V1.0.0
Designed for high-throughput screening and structured job management.
- Multiple input modes: CSV, FASTA, YAML ZIP/folder
- Manifest-driven batch execution
- Input validation and run profiles
CapabilityV1V2Batch
Single protein prediction✓✓✓
Protein–ligand binding✓✓✓
DNA / RNA support—✓—
Templates & custom MSAPartial✓✓
PTMs / constraints / covalent links—✓—
Batch processing——✓
Batch mode
Designed for high-throughput screening and structured job management.
The batch notebook wraps multiple jobs into a manifest-driven system with preflight validation, configurable run profiles, result ranking, and exportable archives.
CSV input
FASTA input
YAML ZIP bundles
YAML folders
Fast / Balanced / Scientific profiles
Skip-complete / retry-failed options
Input builder
Validate job specifications before expensive runs start.
Execution engine
Launch, resume, queue, and monitor multiple jobs in one notebook flow.
Ranking layer
Sort candidates by affinity probability, confidence, pLDDT, pTM, and related metrics.
Archival export
Bundle predictions, manifests, logs, and outputs into a shareable ZIP artifact.
Release highlights
What changed across versions.
Advanced modeling stack
Template upload, explicit chain mapping, covalent bonds, pocket conditioning, contact conditioning, modified residues, custom MSA, DNA/RNA support, and cyclic polymer options.
Public notebook foundation
Initial release with protein–ligand prediction, affinity workflow, visualization, Drive export, and ZIP output packaging.
Automation layer
Multiple predictions, manifest-based execution, ranking, and large-scale run organization for practical screening.
FAQ
Start with the most important questions, then open the full FAQ page.
These answers reflect the current notebook lineup, the batch guide, and the release notes so visitors can quickly find the right workflow.
Do I need a local GPU or CUDA installation?
No. The notebooks are designed around Google Colab, so you can run the main workflow without managing a local CUDA setup.
Which notebook should I start with?
Use the main notebook for general predictions, V2 for advanced modeling, and Batch when you need to screen many jobs at once.
What kinds of inputs are supported?
The repository supports proteins, ligands, templates, custom MSAs, DNA/RNA in V2, and CSV/FASTA/YAML batch manifests.
What outputs can I expect?
Predicted structures, confidence metrics such as pLDDT and PAE, affinity-oriented outputs, visualizations, logs, and exportable ZIP bundles.