Abstract
Recently, it has become possible to obtain multi-omics data from a single cell. Thus, we can now observe the state of
individual cells from a multi-omics perspective, which may provide useful knowledge in the fields of animal science and breeding.
When multi-omics data are obtained at the cellular level, the data is generally compressed to a lower dimension by dimensionality
reduction methods integrating high-dimensional and multi-modal multi-omics data. It is necessary to compare existing methods to
select an appropriate dimensionality reduction method and to develop a new method; however, comparative studies are limited to
date. Thus, in this study, we compared several multiple dimensionality reduction methods using two multi-omics data comprising
gene expression and chromatin accessibility derived from single-cell analysis. The methods included in this study were multiple
co-inertia analysis (MCIA), multi-omics factor analysis,
single-cell aggregation and integration (scAI), Seurat,
principal component analysis, and uniform manifold
approximation and projection. The effectiveness of
dimensionality reduction was evaluated by calculating the silhouette coefficients, which indicate the accuracy of classification of known cell types in the reduced dimensions. We found that
compared with other methods, MCIA and scAI were superior in terms of cell classification but inferior in terms of computation
time. Our results suggested that no method is superior in terms of both accuracy of cell classification and computation time.
