Background Representation learning models have become a mainstay of modern genomics. These models are trained to yield vector representations, or embeddings, of various biological entities, such as cells, genes, individuals, or genomic regions. Recent applications of unsupervised embedding approaches have been shown to learn relationships among genomic regions that define functional elements in a genome. Unsupervised representation learning of genomic regions is free of the supervision from curated metadata and can condense rich biological knowledge from publicly available data to region embeddings. However, there exists no method for evaluating the quality of these embeddings in the absence of metadata, making it difficult to assess the reliability of analyses based on the embeddings, and to tune model training to yield optimal results. Methods To bridge this gap, we propose four evaluation metrics: the cluster tendency test (CTT), the reconstruction test (RCT), the genome distance scaling test (GDST), and the neighborhood preserving test (NPT). The CTT and RCT are statistical methods that evaluate how well region embeddings can be clustered and how much the embeddings can preserve the information contained in training data. The GDST and NPT exploit the biological tendency of regions close in genomic space to have similar biological functions; they measure how much such information is captured by individual region embeddings and a set of region embeddings. Results We demonstrate the utility of these statistical and biological tests for evaluating unsupervised genomic region embeddings and provide guidelines for learning reliable embeddings. Availability Code is available at https://github.com/databio/geniml .

Methods for evaluating unsupervised vector representations of genomic regions