Fotomics: fourier transform-based omics imagification for deep learning-based cell-identity mapping using single-cell omics profiles

Different omics profiles, depending on the underlying technology, encompass measurements of several hundred to several thousand molecules in a biological sample or a cell. This study develops upon the concept of omics imagification as a process of transforming a vector representing these numerical measurements into an image with a one-to-one relationship with the corresponding sample. The proposed imagification process transforms a high-dimensional vector of molecular measurements into a two-dimensional RGB image to enable holistic molecular representation of a biological sample and to improve the classification of different biological phenotypes using automated image recognition methods in computer vision. A transformed image represents 2D coordinates of molecules in a neighbour-embedded space representing molecular abundance and gene intensity. The proposed method was applied to a single-cell RNA sequencing (scRNA-seq) data to imagify gene expression profiles of individual cells. Our results show that a simple convolutional neural network trained on single-cell transcriptomics images accurately classifies diverse cell types outperforming the best-performing scRNA-seq classifiers such as support vector machine and random forest.

Automated summary

This study proposes a method to transform high-dimensional molecular measurements into two-dimensional images, improving the representation and classification of biological samples using automated image recognition methods.