bulktrajblend-trajectory-interpolation
Extend scRNA-seq developmental trajectories with BulkTrajBlend by generating intermediate cells from bulk RNA-seq, training beta-VAE and GNN models, and interpolating missing states.
Security Assessment
About bulktrajblend-trajectory-interpolation
The BulkTrajBlend trajectory interpolation skill is designed to address gaps in single-cell developmental trajectories by leveraging matched bulk RNA-seq data. This skill helps users generate intermediate cell states where missing data exists in developmental pathways, using BulkTrajBlend to extend scRNA-seq trajectories. By employing methods like beta-VAE and Graph Neural Networks (GNN), this tool enables the interpolation of interrupted cell states, providing a more continuous representation of biological processes, even when single-cell data is incomplete or unavailable for certain stages of development.
FAQ
How do I prepare the input data for BulkTrajBlend?
You need to load the reference scRNA-seq AnnData and bulk RNA-seq data. The single-cell data should be processed with tools like `scv.datasets.dentategyrus()` and the bulk RNA-seq counts should be read and harmonized using `ov.utils.read(...)` and `ov.bulk.Matrix_ID_mapping(...)`.
Can I use BulkTrajBlend for data other than PDAC?
Yes, BulkTrajBlend can be applied to any matched bulk RNA-seq and single-cell RNA-seq datasets. The tutorial provided uses PDAC samples as an example, but the tool is versatile and can be adapted to other datasets.
What should I do if the VAE model fails to converge?
If the VAE collapses, you may need to adjust the learning rate (lower it) or reduce the hidden size of the model. This helps prevent high reconstruction loss and improves model performance.
What are the hardware requirements for running BulkTrajBlend?
The skill can be run on systems with GPU support, especially for GNN training, where you need to specify the GPU device using the `gpu` parameter. The bulk RNA-seq and single-cell data processing does not require extensive hardware resources.
Install bulktrajblend-trajectory-interpolation
Quick Setup:
- Copy the skill folder to
.claude/skills/ - Claude will automatically detect and use the skill
Repository
Starlitnightly/omicverse