Selection Sampling from Large Data Sets for Targeted Inference in Mixture Modeling.

Area of Science:

• Computational statistics
• Biomedical data analysis
• Machine learning

Background:

• Markov chain Monte Carlo (MCMC) methods are computationally intensive for large datasets.
• Subsampling techniques are crucial for managing large-scale data analysis.
• Identifying rare cell subtypes in flow cytometry requires efficient analytical approaches.

Purpose of the Study:

• To develop an efficient MCMC approach for analyzing large datasets, particularly for identifying rare cell subtypes.
• To address the computational challenges of scanning entire datasets in MCMC.
• To enhance the precision of inferences for low-probability events in mixture models.

Main Methods:

• A novel MCMC approach utilizing guided selection sampling.
• Sequential Monte Carlo (SMC) strategy for augmenting targeted subsamples.
• Development of a stopping rule for determining optimal subsample size.

Main Results:

• The proposed method effectively targets informative subsamples from large datasets.
• Sequential augmentation improves estimates as the analysis progresses.
• Demonstrated increased resolution in identifying rare cell subtypes using flow cytometry data.

Conclusions:

• The novel MCMC strategy significantly improves computational efficiency for large-scale data analysis.
• The approach enhances the ability to identify and characterize rare cell populations.
• This method offers a powerful tool for complex biological data analysis, especially in flow cytometry.