Topic models for single-cell genomics - BALSAM and DeltaTopic

Key Contributor(s)

• Yichen Zhang: Developer

This is a documentation for the BALSAM and DeltaTopic models, proposed in Zhang et al..

BALSAM, short for Bayesian Latent topic analysis with Sparse Association Matrix, is a Bayesian topic modelling approach to summarize static transcriptome patterns from raw gene expression count data. BALSAM views cells as an admixture of gene topics, and relies on Variational AutoEncoder (VAE) and sparse-inducing priors to learn cell topics and infer the gell-topic relationship.

DeltaTopic, short for Dynamically- Encoded Latent Transcriptomic pattern Analysis by Topic modeling, is a extension of BALSAM model to ascertain common cellular topic space and topic-specific relationships between the unspliced and spliced data.

References

Zhang et al. (2023), Unraveling dynamically-encoded latent transcriptomic patterns in pancreatic cancer cells by topic modelling, preprint.