Using transfer learning from prior reference knowledge to improve the clustering of single-cell RNA-Seq data

Abstract

In many research areas scientists are interested in clustering objects within small datasets while making use of prior knowledge from large reference datasets. We propose a method to apply the machine learning concept of transfer learning to unsupervised clustering problems and show its effectiveness in the field of single-cell RNA sequencing (scRNA-Seq). The goal of scRNA-Seq experiments is often the definition and cataloguing of cell types from the transcriptional output of individual cells. To improve the clustering of small disease- or tissue-specific datasets, for which the identification of rare cell types is often problematic, we propose a transfer learning method to utilize large and well-annotated reference datasets, such as those produced by the Human Cell Atlas. Our approach modifies the dataset of interest while incorporating key information from the larger reference dataset via Non-negative Matrix Factorization (NMF). The modified dataset is subsequently provided to a clustering algorithm. We empirically evaluate the benefits of our approach on simulated scRNA-Seq data as well as on publicly available datasets. Finally, we present results for the analysis of a recently published small dataset and find improved clustering when transferring knowledge from a large reference dataset. Implementations of the method are available at https://github.com/nicococo/scRNA.

Machine learning for single cell genomics data analysis

“…One interesting idea to use complex models on small datasets is to leverage larger, already annotated, datasets to learn the embedding, using techniques from the field of transfer learning or domain adaptation. Embeddings learned by PCA and non-negative matrix factorisation (NMF) on datasets such as the Human Cell Atlas (HCA) have successfully been used in both scATAC-seq [21] and scRNA-seq [22, 23] on new unseen datasets and cell types, as well as used for denoising the new dataset [24]. Similarly the embeddings learned by (demoising) AEs on one dataset, have been shown to be useful on other datasets, both for clustering [25, 26 • , 27, 28] and surface protein prediction [29].…”

Biologically-relevant transfer learning improves transcription factor binding prediction

“…Transfer learning-reusing the information learned from a model developed for one task as the starting point for a model on a second different, but related, task-has been shown to reduce the amount of data required for training while improving overall model performance for diverse applications (reviewed in [13]). In biology, transfer learning has been successful in several areas, including: reconstructing gene regulatory networks [14][15][16]; modeling gene expression from single-cell data [17][18][19][20]; or predicting genomic features, including accessible regions [21], chromatin interactions [22], and TFBSs [23,24].…”

Transfer learning between preclinical models and human tumors identifies a conserved NK cell activation signature in anti-CTLA-4 responsive tumors

“…In this application, we use gene signatures from CoGAPS for projection and transfer learning. Other transfer learning methods have been developed to relate features in a target scRNA-seq dataset to a reference atlas, often relying on non-linear methods for feature identification [77,78]. In contrast to these other approaches, our projectR software is robust for transfer learning from single-cell data (e.g., PCA, clustering, and other forms of linear matrix factorization) and may capture additional features of cell state transitions based upon all of these methodologies [7,30].…”