bulktrajblend-trajectory-interpolation

To fill missing developmental states in a scRNA-seq trajectory, you can use matched bulk RNA-seq to generate intermediate synthetic cells via beta-VAE training and restore trajectory continuity. This method applies graph-based community detection to assign synthetic cells to lineages.

Generating synthetic single cells from bulk RNA-seq data requires training a beta-VAE on harmonised bulk count matrices and a scRNA-seq reference AnnData. The trained model generates intermediate single-cell profiles to bridge gaps in developmental atlases.

Yes, GPU-capable training is required to interpolate scRNA-seq trajectories with bulk counts using deep models like beta-VAE and GNN. Reproducible VAE generation seeds and harmonised gene identifiers between datasets are also required for successful interpolation.

To compute pseudotime and compare topology after trajectory interpolation, run VIA pseudotime analysis and compare PAGA graphs between raw and interpolated data. You can also compute MDE or PCA embeddings to visualise the restored trajectory continuity.

Yes, you can resume trajectory interpolation workflows using saved VAE and GNN checkpoints. The workflow supports checkpoint save and load functionality, allowing you to recover missing lineage states without retraining deep models from scratch.