Models
Data
Benchmarks
Try Models

Configuration Options

Last Updated: January 27, 2025

Reference guide to all workflow parameters and configuration options available in the AI Workspace.

Workflow Configuration

Basic Settings

Workflow Name

  • Purpose: Identify your analysis for later reference
  • Default: Auto-generated timestamp-based name
  • Character limit: 50 characters maximum

Workflow Type

  • Standard: Single-dataset analysis without reference integration
  • Comparison: Integration with CELLxGENE Census reference data
  • Requirements: Comparison requires scVI or TranscriptFormer models

Model Configuration

Model Selection

Choose the computational method for dimensionality reduction:

  • PCA (Principal Component Analysis)
  • scVI (Single-cell Variational Inference)
  • TranscriptFormer (Transcript-level embedding model)

TranscriptFormer Variants

Sapiens
  • Organism: Human only
  • Training: Human-specific pre-training
Exemplar
  • Organism: Human and mouse
  • Training: Multi-organism pre-training
Metazoa
  • Organism: Broad taxonomic support
  • Training: Large-scale multi-species

Classification Settings

Cell Type Prediction

Automatic cell type annotation using trained classifiers:

Availability

  • Models: scVI and TranscriptFormer only
  • Organisms: Human and mouse only

Configuration

  • Enable/Disable: Toggle cell type prediction
  • Default: Disabled (opt-in feature)
  • Processing impact: Adds 5-10 minutes to workflow time

Comparison Workflow Settings

Reference Data Selection

Controls how reference cells are selected from CELLxGENE Census:

Enabled

  • Method: Nearest neighbor search
  • Selection: Up to 1M most similar reference cells
  • Processing time: Longer due to similarity computation

Disabled

  • Method: Random sampling
  • Selection: Up to 250k random reference cells
  • Processing time: Faster processing

Tissue Filters

Available Filters: All tissue in CELLxGENE Census.

Multiple Selection

  • Select multiple tissue types simultaneously
  • Logical OR relationship (cells from any selected tissue)

Job Limits

  • Concurrent jobs: Maximum 3 jobs per user
  • Result retention: 24 hours after completion
  • File size: Maximum 5GB input files

Parameter Validation

Compatibility Checks

System automatically validates:

  • Model-organism compatibility: Ensures supported combinations
  • Workflow-model compatibility: Verifies comparison workflow requirements
  • Classification requirements: Checks model and organism support