(Table1), which explicitly take into account all scRNA-seq-specific
confounding factors.
Unlike the DEG analysis, the identification of alternative splic-
ing (AS) events or exons is much more challenging, owing to the
unique properties of scRNA-seq data as low coverage, 3^0 bias, and
technical noise. Although the technical noise is optimistically
resolvable through computational methods described above,
another two peculiarities make AS study problematic. So far, there
is only one reported effort aiming at AS detection based on scRNA-
seq reads, called SingleSplice [123]. However, this method only
enables the detection of partially AS fragments rather than the full-
length AS transcripts that are easily obtained from bulk RNA-seq
reads. Additionally, SingleSplice is limited to tests for variability in
groups of cells, rather than in really individual cell. Alternatively, to
circumvent the detection of different types of AS event directly,
methods that characterize the transcripts that display differences in
exon or isoform usage have been reported [124, 146]. As in studies
of differential expression, tools for identifying differentially
expressed exons [147, 148] can be used in these cases. Therefore,
AS detection based on scRNA-seq data is still problematic largely
due to its low coverage and 3^0 bias, which gives the great chances
for both the technical improvement and computational method
development. Our group are now working on the possibility of 5^0
coverage imputation for single-cell sequencing, which holds the
promise to overcome the 3^0 bias computationally for AS or more
other analyses which are commonly used for bulk RNA-seq data.2.4.3 Modeling
Transcriptional Dynamics
Growing evidence suggests that genes are not transcribed consis-
tently but rather undergo highly dynamic expression patterns
across a population of cells. scRNA-seq can be used to explore
transcriptional kinetics of cells but still of challenge without prior
knowledge of the underlying cell types. Nevertheless, unlike the
population-averaged data from bulk RNA-seq data, scRNA-seq can
characterize diversity in transcription between individual cells to the
acceptable extent [149, 150]. Identification of the highly dynamic
genes requires the application of statistical approaches that account
for technical sources of visibilities, such that biological variability in
gene expression levels can be quantified accurately and sensitively.
Additionally, the high variability of gene expression can also be
caused by confounding factors that are not accounted for, such as
the cell cycle [139]. One approach is to compute the coefficient of
variation for each gene across the population of cells under study
and to rank the genes accordingly. Unfortunately, technical varia-
bility, which is intrinsic to the experimental protocol and not asso-
ciated with genuine biological variability, is greater for lowly
expressed genes than for highly expressed genes [1]. Consequently,
a null estimate of the expected technical variability needs to beApplications of Single-Cell Sequencing for Multiomics 357