User Guide

torchcell is a provenance-first experimental-data ontology and dataset substrate for the yeast (Saccharomyces cerevisiae) virtual cell. Every record is a typed experiment, genotype x environment -> phenotype, paired with the reference it is measured against and traceable to the source file it was parsed from. Records are built into LMDB-backed dataset loaders, served together as a Neo4j knowledge graph, and queried back into PyTorch Geometric datasets for model training.

• Installation
  • Create the environment
  • Environment variables
  • Verify the install
• Quickstart
  • Load the reference genome
  • Load an experiment dataset
  • Query the served knowledge graph
  • Inspect a stored record
• Data model
  • The experiment record
  • Verification levels
  • Provenance of hand-entered values
• Datasets
  • How a dataset is built
  • Raw files and their provenance
  • Supported datasets
• Knowledge graph
  • Schema
  • Writing a query
  • Choosing a release
  • Converters
  • Changing what the graph serves
• Contributing
  • Branches, pull requests and landing
  • Tests
  • Pre-commit hooks
  • Commit messages and releases
  • Adding a dataset page
• Downloading datasets
  • Swagger
  • Raw, genomes and objects tiers
  • curl
  • Environment variables
  • From Python

Released versions of the served knowledge graph, and which torchcell versions they correspond to, are listed under Database releases.