Spectrogram-Based Imagification Applying Deep Learning on Omics Data

In the field of bioinformatics, 'omics' refers to the comprehensive study and analysis of various biological molecules such as genes, proteins and metabolites, to understand their roles, relationships and functions within an organism. In recent years, the proliferation of omics data and advancements in machine learning have transformed the field of bioinformatics. Although deep neural networks such as convolutional neural networks have been widely applied in fields such as computer vision and biomedical image computing, they have been less applied to biological data such as omics. One of the reasons for this is that omics data is generally one-dimensional (1D), whereas convolutional neural networks need input in the form of a two-dimensional (2D) matrix for enhanced feature extraction compared to 1D CNN networks. In this work, we propose a novel spectrogram-based method to convert high-dimensional omics data into 2D images, which facilitates the application of convolutional neural networks for downstream tasks such as cancer classification. We evaluate the proposed method against state-of-the-art techniques such as DeepInsight and Fotomics across five diverse datasets. Our proposed method not only shows its effectiveness by performing well across standard evaluation metrics but also is computationally less demanding compared to existing methods. We believe that the proposed methodology, when applied to omics data by transforming 1D data into 2D images, can help leverage the benefit of deep learning models, in turn benefitting various bioinformatics applications such as gene-gene prediction, drug discovery and identification of potential biomarkers for diseases.