Fast clustering and cell-type annotation of scATAC data using pre-trained embeddings

Abstract

MotivationData from the single-cell assay for transposase-accessible chromatin using sequencing (scATAC-seq) is now widely available. One major computational challenge is dealing with high dimensionality and inherent sparsity, which is typically addressed by producing lower-dimensional representations of single cells for downstream clustering tasks. Current approaches produce such individual cell embeddings directly through a one-step learning process. Here, we propose an alternative approach by building embedding models pre-trained on reference data. We argue that this provides a more flexible analysis workflow that also has computational performance advantages through transfer learning.ResultsWe implemented our approach in scEmbed, an unsupervised machine learning framework that learns low-dimensional embeddings of genomic regulatory regions to represent and analyze scATAC-seq data. scEmbed is competitive with alternative scATAC embedding approaches in terms of clustering ability and has the advantage of learning patterns of region co-occurrence that can be transferred to other, unseen datasets. Moreover, pre-trained models on reference data can be exploited to build fast and accurate cell-type annotation systems without the need for other data modalities. scEmbed is implemented in Python and it is available to download from GitHub. We also make our pre-trained models available on huggingface for public use.AvailabilityscEmbed is open source and available athttps://github.com/databio/geniml. Pre-trained models from this work can be obtained on huggingface:https://huggingface.co/databio.

Joint Representation Learning for Retrieval and Annotation of Genomic Interval Sets

“…Our approach leverages recent advances in neural embedding methods and the growing corpus of epigenome data to tie natural language to genomic intervals. Neural embedding approaches show great promise for a variety of biological applications [ 22 , 23 , 24 , 25 , 26 , 27 , 28 ]. In particular, the StarSpace neural embedding approach has been recently used to learn representations of cancer mutational signatures and has shown to be resource-efficient, flexible, and scalable [ 29 ].…”

Methods for evaluating unsupervised vector representations of genomic regions

“…We previously demonstrated the value of region set embeddings. We reason that the underlying region embeddings could be useful independently; for example, we can infer the function of an unknown region based on its closeness to other known regions in the embedding space, or use them for fast annotation and clustering on single-cell data (17). To develop such methods and concepts further, we first sought a way to evaluate region embeddings independently and objectively.…”