A1122
Title: Compositional Data Analysis with Image generator methods
Authors: Je seok Lee - Kyungpook National University (Korea, South) [presenting]
Byungwon Kim - Kyungpook National University (Korea, South)
Abstract: Next Generation Sequencing (NGS), exemplified by 16S rRNA sequencing, is a widely used method for interpreting microbiome data. Although NGS microbiome data possess the characteristics of compositional data, past research has often ignored this and used standard normalization procedures for analysis. Recently, there has been work in Compositional Data Analysis (CoDA) that takes into account the compositional nature of NGS microbiome data, and this study follows that approach. In studying NGS microbiome data through the CoDA approach, most have focused on transformations, in particular, log-ratio typed transformations which move the sample space to Euclidean vector space. In this study, we applied image generator methods after transforming the data to Euclidean vector space. Image generator methods aim to use Convolutional Neural Networks (CNNs)one of the leading models in machine learning for image classification non-image data. This approach is particularly strong in handling irregular and extensive datasets. Our results demonstrate the utility of CoDA approaches and the effectiveness of image generator methods for reinterpreting NGS microbiome data.
