Deep Learning for Raman Spectroscopy analysis of SARS-CoV-2 RNA

Raman Spectroscopy has proved to be an effective technique for investigating the chemical composition of organic samples. In several contexts Raman Spectra show clear indicators of specific bonds and simple molecular structures. However, in the case of large polymers such as RNA, spectra appear to be influenced by the primary and secondary structure of the molecules, thereby posing a significant challenge to sample analysis. In this work we propose a Deep Learning approach for analyzing data obtained from Sars-CoV-2 RNA. We employ Convolutional Neural Networks to perform classification and regression tasks involving Raman spectra. We introduce Selective State Space models and test their performance on a classification task involving nucleotide sequences. We discuss two possible applications of Self-supervised learning: an Adversarial model for generating new spectra and a rudimentary Multimodal model for embedding spectra and nucleotide sequences in a shared space.