A0340
Title: Dimensionality reduction for multi-omics data using the Freeman-Tukey transformation
Authors: Hiroshi Kobayashi - Doshisha University (Japan) [presenting]
Masaaki Okabe - Doshisha University (Japan)
Hiroshi Yadohisa - Doshisha University (Japan)
Abstract: The data integration method is useful for understanding the complex patterns between multiple datasets. When multiple datasets contain complementary information, data integration can remove noise and discover common structures. The Co-Inertia Analysis (CIA) method describes the relationships between multiple datasets by maximizing the covariance between them. Therefore, it captures the structure associated with multiple datasets and enables low-dimensional visualization without losing the unique information contained in each dataset. When applied to overdispersed data or highly sparse count data, CIA often requires a combination with logarithmic transformation. However, dimension reduction through logarithmic transformation may distort the structure of the data. We propose CIA, using the Freeman-Tukey transformation. This technique is able to better capture low-dimensional structures that preserve the structure of the original count data. It also enables dimension reduction of multiple datasets without compromising the biological structure, while integrating real multi-omics data. Moreover, it allows for downstream analysis that accurately captures the inherent biological features, resulting in deeper insights into complex biological characteristics.
