Batch effects and the effective design of single-cell gene expression studies.
Tung PY, Blischak JD, Hsiao CJ, Knowles DA, Burnett JE, Pritchard JK, Gilad Y. · 2017
Read Batch effects and the effective design of single-cell gene expression studies. by Tung PY, Blischak JD, Hsiao CJ, Knowles DA, Burnett JE, Pritchard JK, Gilad Y. (2017) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc5206706.txt.
Introduction
Po-Yuan Tung, John D. Blischak, Chiaowen Joyce Hsiao, David A. Knowles, Jonathan E. Burnett, Jonathan K. Pritchard, and Yoav Gilad published a methods study in Scientific Reports in 2017 on noise in single-cell RNA sequencing. The work asks where the unwanted variation in these experiments comes from and how study design can limit it.
Working on the Fluidigm C1 platform, the group ran three replicates from three human stem cell lines and added molecular tags meant to correct for copying bias. Genotype drove most of the real differences, yet repeat runs of the same material still varied a lot. Counts from the molecular tags also shifted with both biology and technique, so they are not a clean measure of gene activity. The paper closes with a practical design framework for future single-cell studies.
Catalog record
| CREATOR | Tung PY, Blischak JD, Hsiao CJ, Knowles DA, Burnett JE, Pritchard JK, Gilad Y. |
|---|---|
| YEAR | 2017 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 62 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Batch effects and the effective design of single-cell gene expression studies By Po-Yuan Tung, John D. Blischak, Chiaowen Joyce Hsiao, David A. Knowles, Jonathan E. Burnett, Jonathan K. Pritchard, Yoav Gilad Published in: Scientific Reports DOI: 10.1038/srep39921 ABSTRACT Single-cell RNA sequencing (scRNA-seq) can be used to characterize variation in gene expression levels at high resolution. However, the sources of experimental noise in scRNA-seq are not yet well understood. We investigated the technical variation associated with sample processing using the single-cell Fluidigm C1 platform. To do so, we processed three C1 replicates from three human induced pluripotent stem cell (iPSC) lines. We added unique molecular identifiers (UMIs) to all samples, to account for amplification bias. We found that the major source of variation in the gene expression data was driven by genotype, but we also observed substantial variation between the technical replicates. We observed that the conversion of reads to molecules using the UMIs was impacted by both biological and technical variation, indicating that UMI counts are not an unbiased estimator of gene expression levels. Based on our results, we suggest a framework for effective scRNA-seq studies. Single-cell genomic technologies can be used to study the regulation of gene expression at unprecedented resolution12. Using single-cell gene expression data, we can begin to effectively characterize and classify individual cell types and cell states, develop a better understanding of gene regulatory threshold effects in response to treatments or stress, and address a large number of outstanding questions that pertain to the regulation of noise and robustness of gene expression programs. Indeed, single cell gene expression data have already been used to study and provide unique insight into a wide range of research topics, including differentiation and tissue development345, the innate immune response67, and pharmacogenomics89. Yet, there are a number of outstanding challenges that arose in parallel with the application of single cell technology10. A fundamental difficulty, for instance, is the presence of inevitable technical variability introduced during sample processing steps, including but not limited to the conditions of mRNA capture from a single cell, amplification bias, sequencing depth, and variation in pipetting accuracy. These (and other sources of error) may not be unique to single cell technologies, but in the context of studies where each sample corresponds to a single cell, and is thus processed as a single unrepeatable batch, these technical considerations make the analysis of biological variability across single cells particularly challenging. To better account for technical variability in scRNA-seq experiments, it has become common to add spike-in RNA standards of known abundance to the endogenous samples1112. The most commonly used spike-in was developed by the External RNA Controls Consortium (ERCC)13; comprising of a set of 96 RNA controls of varying length and GC content. A number of single cell studies focusing on analyzing technical variability based on ERCC spike-in controls have been reported11121415. However, one principle problem with spike-ins is that they do not ‘experience’ all processing steps that the endogenous sample is subjected to. For that reason, it is unknown to what extent the spike-ins can faithfully reflect the error that is being accumulated during the entire sample processing procedure, either within or across batches. In particular, amplification bias, which is assumed to be gene-specific, cannot be addressed by spike-in normalization approaches. To address challenges related to the efficiency and uniformity with which mRNA molecules are amplified and sequenced in single cells, unique molecule identifiers (UMIs) were introduced to single cell sample processing16171819. The rationale is that by counting molecules rather than the number of amplified sequencing reads, one can account for biases related to amplification, and obtain more accurate estimates of gene expression levels71220. It is assumed that most sources of variation in single cell gene expression studies can be accounted for by using the combination of UMIs and a spike-in based standardization1520. Nevertheless, though molecule counts, as opposed to sequencing read counts, are associated with substantially reduced levels of technical variability, a non-negligible proportion of experimental error remains unexplained. There are a few common platforms in use for scRNA-seq. The automated C1 microfluidic platform (Fluidigm), while more expensive per sample, has been shown to confer several advantages over platforms that make use of droplets to capture single cells321. In particular, smaller samples can be processed using the C1 (when cell numbers are limiting), and the C1 capture efficiency of genes (and RNA molecules) is markedly higher. Notably, in the context of this study, the C1 system also allows for direct confirmation of single cell capture events, in contrast to most other microfluidic-based approaches322. One of the biggest limitations of using the C1 system, however, is that single cell capture and preparation from different conditions are fully independent23. Consequently, multiple replicates of C1 collections from the same biological condition are necessary to facilitate estimation of technical variability even with the presence of ERCC spike-in controls10. To our knowledge, to date, no study has been purposely conducted to assess the technical variability across batches on the C1 platform. To address this gap, we collected scRNA-seq data from induced pluripotent stem cell (iPSC) lines of three Yoruba individuals (abbreviation: YRI) using C1 microfluidic plates. Specifically, we performed three independent C1 collections per each individual to disentangle batch effects from the biological covariate of interest, which, in this case, is the difference between individuals. Both ERCC spike-in controls and UMIs were included in our sample processing. With these data, we were able to elucidate technical variability both within and between C1 batches and thus provide a deep characterization of cell-to-cell variation in gene expression levels across individuals. RESULTS Study design and quality control We collected single cell RNA-seq (scRNA-seq) data from three YRI iPSC lines using the Fluidigm C1 microfluidic system followed by sequencing. We added ERCC spike-in controls to each sample, and used 5-bp random sequence UMIs to allow for the direct quantification of mRNA molecule numbers. For each of the YRI lines, we performed three independent C1 collections; each replicate was accompanied by processing of a matching bulk sample using the same reagents. This study design (Fig. 1a and Supplementary Table S1) allows us to estimate error and variability associated with the technical processing of the samples, independently from the biological variation across single cells of different individuals. We were also able to estimate how well scRNA-seq data can recapitulate the RNA-seq results from population bulk samples. In what follows, we describe data as originating from different samples when we refer to data from distinct wells of each C1 collection. Generally, each sample corresponds to a single cell. In turn, we describe data as originating from different replicates when we refer to all samples from a given C1 collection, and from different individuals when we refer to data from all samples and replicates of a given genetically distinct iPSC line. We obtained an average of 6.3 +/− 2.1 million sequencing reads per sample (range 0.4–11.2 million reads). We processed the sequencing reads using a standard alignment approach (see Methods) and performed multiple quality control analyses. As a first step, we estimated the proportion of ERCC spike-in reads from each sample. We found that, across samples, sequencing reads from practically all samples of the second replicate of individual NA19098 included