The abstract is read at the publisher; the summary is JClub's.
TMCells identifies high-quality metacells in single-cell omics by learning latent topic distributions with a Gaussian-mixture prior, capturing continuous state transitions and uncertainty.
- Why it matters: Existing metacell methods often lack biological interpretability, use deterministic assignments that miss continuous transitions, and can discard signals during feature selection or dimensionality reduction.
- What they did: TMCells uses a probabilistic generative framework with topic modeling and a Gaussian mixture model prior for topic distributions, learning topic representations and metacell partitions end-to-end.
- The result: The learned metacells reduce technical noise while preserving biological heterogeneity, improving downstream differential expression, rare cell type identification, and trajectory inference.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.