TphPMF: A microbiome data imputation method using hierarchical Bayesian Probabilistic Matrix Factorization.

Area of Science:

• Microbiome research
• Bioinformatics
• Machine learning

Background:

• Data sparsity is a major challenge in microbiome research, affecting analysis accuracy and biological insights.
• Existing imputation methods struggle to reliably address missing data in complex microbial datasets.

Purpose of the Study:

• To introduce TphPMF, a machine learning approach for microbiome data imputation.
• To evaluate TphPMF's performance against existing methods and its impact on downstream analyses.

Main Methods:

• Probabilistic Matrix Factorization (PMF) with phylogenetic priors.
• Bayesian inference to predict non-biological zeros in microbiome data.
• TphPMF leverages microbial phylogenetic relationships for improved imputation.

Main Results:

• TphPMF significantly outperforms current methods in recovering missing taxon abundances.
• Imputation with TphPMF enhances differential abundance analysis, especially with DESeq2-phyloseq.
• TphPMF improves disease prediction accuracy for type 2 diabetes and colorectal cancer datasets.

Conclusions:

• TphPMF offers a robust solution for microbiome data sparsity.
• This method enhances the reliability of microbiome data analysis and disease association studies.
• TphPMF facilitates more accurate biological discoveries from sparse microbiome data.