Overview

ST-CNVBench provides a public benchmark controller for CNV inference on spatial transcriptomics datasets. The workflow is organized into three main stages:

prep -> run -> eval

Design Goals

• standardize heterogeneous spatial transcriptomics inputs into one benchmark-ready layout
• run multiple CNV inference methods through one shared controller
• evaluate results with task-specific loaders and metrics
• keep the public surface config-driven and method-extensible

Pipeline Stages

Dataset Preparation

The prep stage standardizes raw spatial transcriptomics inputs into a common output bundle used by downstream wrappers. Supported public input layouts are documented in Dataset Preparation.

Model Run

The run stage executes one or more CNV inference methods against prepared datasets using conda, docker, or apptainer runtime modes. See Model Run for the public model config structure.

Evaluation

The eval stage loads method outputs through public adapters and runs selected benchmark tasks such as cnv_profile or tumor_normal. See Evaluation for available tasks and GT requirements.

Public Examples

The docs include three task-focused examples:

• CNV Profile Task Example
• Tumor-Normal Classification Task Example
• Subclone Identification Task Example

For the packaged cSCC walkthrough, go to Quickstart Demo And Expected Outputs.