Query to reference single-cell integration with transfer learning

Abstract

Large single-cell atlases are now routinely generated with the aim of serving as reference to analyse future smaller-scale studies. Yet, learning from reference data is complicated by batch effects between datasets, limited availability of computational resources, and sharing restrictions on raw data. Leveraging advances in machine learning, we propose a deep learning strategy to map query datasets on top of a reference called single-cell architectural surgery (scArches, https://github.com/theislab/scarches). It uses transfer learning and parameter optimization to enable efficient, decentralized, iterative reference building, and the contextualization of new datasets with existing references without sharing raw data. Using examples from mouse brain, pancreas, and whole organism atlases, we showcase that scArches preserves nuanced biological state information while removing batch effects in the data, despite using four orders of magnitude fewer parameters compared to de novo integration. To demonstrate mapping disease variation, we show that scArches preserves detailed COVID-19 disease variation upon reference mapping, enabling discovery of new cell identities that are unseen during training. We envision our method to facilitate collaborative projects by enabling the iterative construction, updating, sharing, and efficient use of reference atlases.

Integrated analysis of multimodal single-cell data

“…For example, cells that were annotated as regulatory T cells expressed CD25 + in the CITE-seq data, and we observed similar results for MAIT cells (CD161 + ), memory (CD45RA -CD45RO + ) and naive (CD45RA + CD45RO -) T cells, and circulating ILC (CD117 + CD25 + , Supplementary Figure 8). We benchmarked our method against scArches, a recently developed method for mapping scRNA-seq queries to reference datasets [55] and observed that our approach yielded substantial improvements in accuracy and performance ( Figure 7A, B; Supplementary Figure 8).…”

A Harmonized Atlas of Spinal Cord Cell Types and Their Computational Classification

“…Theis and colleagues propose a method called single-cell architectural surgery that uses transfer learning to map query datasets onto a reference, simultaneously contextualizing the query while updating the reference. This allows for decentralized reference building without the sharing of raw data, which could further increase effectiveness of neural network-based classifiers 20 . There are several notable limitations to this study and to single cell transcriptomics in general.…”

scMC learns biological variation through the alignment of multiple single-cell genomics datasets

“…As expected, the cells with the same labels from the third batch are correctly placed onto this shared UMAP space (Additional file 2: Figure S14). However, projecting cells from very different sources with strong batch effects into a shared low-dimensional space may require a more advanced transfer learning approach, such as scArches [41].…”