Advanced Use

• Here we provide a step-by-step guide to follow the TransBrain construction pipeline.

• If you have more specific needs (such as constructing mapping matrices based on your own data or atlases), this tutorial will help you.

• On each page you can find a brief introduction of the data and methods, for detailed description please refer to our preprint.

• Due to the large data size, the dataset is not included in the GitHub repository. If needed, you can download it from the link in pipeline/datasets/README.md and place it in the same directory.

1. Spatial Transcriptomic Matching

• We integrated complementary human transcriptomic datasets, including microarray data and large-scale single-nucleus RNA sequencing data.

• A detached deep neural network model was trained on the integrated transcriptomic data to learn region-specific latent embeddings that are generalizable across species.

2. Graph-based Random Walk

• A heterogeneous graph was constructed to connect brain regions within and across species.

• Latent embeddings capturing integrated transcriptomic, anatomical, and connectivity information were generated.

3. Bidirectional Mapping

• Dual-regression method was employed to map phenotypes cross species, using the latent embeddings defined previously.

Steps

• Transcriptional Data Integration
• DNN Model Training
• Graph-based Random Walk
• Bidirectional Mapping