/CovidAutoEncoder

Created and tested various autoencoder architectures (Convolutional, Variational, Linear) to learn latent representations of SARS-Cov-2 genomes and most importantly: the latent differences between the variant genomes.

Primary LanguageJupyter NotebookMIT LicenseMIT

CovidAutoEncoder

Mihir Bafna, Vikranth Keerthipati, Ruhan Ponnada

CS 4641 Final Project (its not that deep)

Overview

Created various autoencoder architectures (Convolutional, Variational, Linear) to learn latent representations of SARS-Cov-2 genomes and most importantly, the latent differences between the variant genomes.

We trained the autoencoder on more prevalent strains (Alpha and Delta). Once the model was capable of learning lower dimensional latent representations without losing information, we encoded Omicron genomes into these representations and were able to clearly distinguish the variants with a simple PCA.