Post-Graduate Stories

Ismaïla Ba Explores “Zero-Inflation in Multinomial Principal Component Analysis for Microbiome Data”

In this project, Ismaïla will develop novel statistical methodology related to zero-inflated multinomial principal component analysis and will explore options for fitting the model through variational Bayes methods.

Program: CANSSI Distinguished Postdoctoral Fellowship
Region: National
Date: 2022–2024

Project Focus Areas

Dimension reduction techniques are among the most essential analytical tools in the statistical analysis of genomic data. Generalized principal component analysis is an extension to standard principal component analysis (PCA) for non-Gaussian data and has been used in the analysis of data from genomic platforms such as single-cell sequencing and microbial/metagenomic sequencing. In particular, multinomial PCA has been adapted for use in these contexts. In microbiome data, however, there is often an abundance of zero counts, which is not accounted for in the multinomial PCA framework. In this project, Ismaïla will develop novel statistical methodology related to zero-inflated multinomial PCA and will explore options for fitting the model through variational Bayes methods.

A common element of data arising from next-generation sequencing technologies is the presence of unmeasured factors affecting the observed data distributions. Such factors can be detected in an unsupervised manner using generalized Principal Component Analysis (gPCA). Ismaïla will perform methodological work to integrate the zero-inflation framework from previous research into the multinomial gPCA model. Specifically, he will first extend the gPCA framework to a zero-inflated generalized Dirichlet multinomial model.

However, the large number of parameters that need to be estimated in gPCA can lead to computational challenges. For this reason, Ismaïla will study and implement a variational EM algorithm for estimating the parameters of the gPCA model. A successful implementation of a variational EM algorithm will make this model appropriate for large-scale genomic studies.

The impact of this project will be a new technique for performing zero-inflated multinomial PCA in microbiome sequencing data. An R package will be designed for use by researchers in the biomedical sciences with a limited quantitative background. To the applicants’ knowledge, there is no such tool implementing this specific model available for biomedical researchers.

Getting to Know Ismaïla

Ismaïla Ba received his PhD in statistics from the Université du Québec à Montréal in 2022.

His long-term research interests include high-dimensional regression, feature screening and selection procedure, and development of cutting-edge machine learning methods with the goal of developing skills and knowledge for a career in the development of statistical methods and computational software for the analysis of high-dimensional data arising from a number of domains.

Ismaïla has a personal interest in science communication and would like to continue to take advantage of opportunities to exercise these skills by penning articles for blogs and online magazine and by presenting his research to non-expert audiences during his postdoctoral term.

As a recipient of the CANSSI Distinguished Postdoctoral Fellowship, Ismaïla will work under the supervision of Professor Kevin McGregor at York University and Professor Maxime Turgeon at the University of Manitoba.