This article presents significant enhancements to the Original Bayesian GPLVM (OBGPLVM) model for analyzing single-cell RNA sequencing data. By implementing pre-processing techniques and adopting a negative binomial likelihood, this modified model achieves better performance. It emphasizes normalizing library sizes and incorporating biological factors such as batch and cell-cycle information. Results demonstrate that each modification is crucial for improving the latent space's consistency with biological relevance, making the model not only more effective but also relevant for biological interpretations. Comparisons with existing models like SCVI affirm its competitive standing in the field.
In this study, we introduce modifications to the baseline Bayesian GPLVM model, demonstrating that pre-processing, likelihood adaptation, and additional technical factors significantly enhance performance.
Our experimental results indicate that each component of the modified model is critical to achieving better performance, overcoming limitations posed by the Original Bayesian GPLVM.
The consistency of the latent space with biological factors underscores the importance of our modifications, making the model not only a robust analytical tool but also biologically interpretable.
By modifying the way we process data and adapt our likelihood, we provide evidence that these changesâtheoretical and practical considerationsâlead to substantial improvements over traditional methods.
#bayesian-gplvm #single-cell-rna-sequencing #statistical-models #data-preprocessing #machine-learning
Collection
[
|
...
]