A self-supervised deep learning method for data-efficient training in genomics

Abstract While deep learning is frequently applied in bioinformatics, it is mostly limited to problems where huge amounts of labeled data are present to train a classifier in a supervised manner. Here, we introduce Self-GenomeNet – a method that utilizes unlabeled genomic data to address the challenge of limited data availability through self-training, outperforming the standard supervised training, even when using ~10 times less labeled data.