Skip to content

tsp-maker

Create TSP (TSL Structure Package) datasets from protein structure predictions.

tsp-maker is a command-line tool that converts outputs from structure prediction tools (AlphaFold2, AlphaFold3, Boltz2) into the TSP format for distribution via Zenodo.

What is TSP?

TSP (TSL Structure Package) is a data standard for distributing protein structure prediction datasets. It extends the Frictionless Data Package specification for structural biology.

A TSP package contains:

  • metadata.parquet - Per-structure statistics and annotations
  • structures/ - Batched structure archives (PDB/mmCIF)
  • predictions/scores.parquet - Prediction confidence scores
  • predictions/pae/ - PAE (Predicted Aligned Error) matrices
  • datapackage.json - Package manifest

TSP packages are consumed by the tslstructures R package.

Workflow Overview

┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│   Predictor     │     │  Intermediate   │     │      TSP        │
│    Outputs      │────▶│     Format      │────▶│    Package      │
│  (AF2/AF3/BZ2)  │     │                 │     │                 │
└─────────────────┘     └─────────────────┘     └─────────────────┘
        │                       │                       │
   tsp-maker parse         tsp-maker build        tsp-maker upload
                                                        │
                                                        ▼
                                                 ┌─────────────────┐
                                                 │     Zenodo      │
                                                 │   tsl-structures│
                                                 │    community    │
                                                 └─────────────────┘

Quick Example

# Parse AlphaFold3 outputs
tsp-maker parse af3 /data/af3_predictions /intermediate

# Build TSP package
tsp-maker build /intermediate /my-dataset \
    --name my-structures \
    --title "My Structure Dataset"

# Upload to Zenodo
tsp-maker upload /my-dataset --token $ZENODO_TOKEN --publish

Features

  • Multi-predictor support - AlphaFold2, AlphaFold3, Boltz2
  • Automatic batching - Structures grouped into manageable archives
  • PAE extraction - Full PAE matrices preserved
  • Zenodo integration - Direct upload with metadata
  • Validation - Check TSP package conformance

Getting Started

  1. Install tsp-maker
  2. Follow the Quick Start guide
  3. Read the command reference