SpaceTravLR is a computational framework for modeling how cells extend their transcriptome beyond their own cytoplasm. SpaceTravLR aims to push spatial analyses at single cell resolution from descriptive correlations toward functional mechanistic insights.

The integration of in-silico perturbation screening with spatial context allows us to infer how the combination of gene-gene interactions rewire signaling to define distinct cellular environments.

We leverage spatial transcriptomics data to uncover functional microniches - tissue regions where local differences in the cellular microenvironment drive divergent cell fates or states.

I want to 🧞️¶

Quickstart¶

Install me¶

curl -fsSL https://tinyurl.com/spacetravlr/scripts/install.sh | sh

Run me¶

spacetravlr --h5ad /path/to/adata.h5ad --output-dir /path/to/outputdir

Join me¶

spacetravlr --join-output-dir  /path/to/outputdir --plain

For example, this slurm job will coordinate multiple workers on the same output/adata

#!/bin/bash
#SBATCH --partition=preempt
#SBATCH --job-name=SpaceTravLR
#SBATCH --mem=300G
#SBATCH --output=/dev/null
#SBATCH --error=/dev/null
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=64
#SBATCH --cluster=gpu
#SBATCH --gres=gpu:1
#SBATCH --time=1-00:00:00

spacetravlr --join-output-dir /path/to/outputdir --plain

Analyze me¶

Here we collect the regulatory landscape by looking at the learned gene-gene interactions across all feather files.

spacetravlr collect-interactions \
  --run-toml /path/to/outputdir/spacetravlr_run_repro.toml

Here we generate functional microniches from the feather files by applying the Leiden algorithm to the learned beta coefficients directly.

spacetravlr get-microniches \
  --run-toml /path/to/outputdir/spacetravlr_run_repro.toml

Point an Ai agent at me¶

SpaceTravLR publishes an llms.txt index and a self-contained llms-full.txt reference following the llmstxt.org convention. Give either URL to Claude, Cursor, ChatGPT, or any coding agent and it will know the full capability surface, every CLI flag, the config schema, and the output formats.

https://spacetravlr-rust.readthedocs.io/en/latest/llms-full.txt

Training time estimate¶

The widget below provides a rough estimate of how long training SpaceTravLR on your dataset will take. This was empirically estimated using multiple runs across datasets on a100, l40s and rtx6k GPUs.

\(s_{\mathrm{gpu}}\) is a GPU model specific coefficient representing the speedup relative to only using the CPU.

cnn_max_cells allows the CNN to smartly subsample the dataset as the training converges. Higher values means later epochs see fewer and fewer cells, focusing on tissue region where the residual errors are higher.