The summary of Dr. Sascha Ott's presentation from the Jun 11-12th 2019 event Data-driven systems medicine at Cardiff University Brain Research Imaging Centre.
2. How does Drop-seq work?
mRNA CAPTURE
Uniquely barcoded
mRNA capture beads
Cells
Cells
Oil
Oil
sample loop
to minimise
bead
damage.
The oil, cells and mRNA capture beads are
pumped through the microfluidic chip.
The two aqueous streams of barcoded mRNA
capture beads and cells mix together less than a
millisecond before the microfluidic junction and
are then encapsulated in droplets.
The lysis buffer in which beads are resuspended
breaks the cells open and the mRNA is captured
on the bead. Following mRNA capture in
droplets, the emulsion that comes off the chip is
4. Data used and key questions
Endometrial data - recurrent miscarriages:
Prof. Jan Brosens
Prof. Siobhan Quenby
5. Data used and key questions
Endometrial data - recurrent miscarriages:
- Fresh patient samples
processed using Drop-seq
Prof. Jan Brosens
Prof. Siobhan Quenby
6. Data used and key questions
Endometrial data - recurrent miscarriages:
- Fresh patient samples
processed using Drop-seq
- Aim to identify differences
between RPL patients and controls
Prof. Jan Brosens
Prof. Siobhan Quenby
7. Data used and key questions
Endometrial data - recurrent miscarriages:
Pancreatic data - pancreas in pregnancy:
- Fresh patient samples
processed using Drop-seq
- Aim to identify differences
between RPL patients and controls
Dr Mike Khan
Prof. Jan Brosens
Prof. Siobhan Quenby
8. Data used and key questions
Endometrial data - recurrent miscarriages:
Pancreatic data - pancreas in pregnancy:
- Fresh patient samples
processed using Drop-seq
- Aim to identify differences
between RPL patients and controls
- Pancreatic islet samples from pregnant
mice processed using Drop-seq
Dr Mike Khan
Prof. Jan Brosens
Prof. Siobhan Quenby
9. Data used and key questions
Endometrial data - recurrent miscarriages:
Pancreatic data - pancreas in pregnancy:
- Fresh patient samples
processed using Drop-seq
- Aim to identify differences
between RPL patients and controls
- Aim to identify mechanisms of pancreatic
adaptation during pregnancy
- Pancreatic islet samples from pregnant
mice processed using Drop-seq
Dr Mike Khan
Prof. Jan Brosens
Prof. Siobhan Quenby
10. Computational Problem
β20
0
20
β20 0 20
tSNE_1
tSNE_2
Endothelial
Epithelial β state 1
Epithelial β state 2
NK
Stroma β state 1
Stroma β state 2
Stroma β state 3
Stroma β state 4
Unknown
17. JACC the Ripper
tailored specifically for scRNA-seq
data
0.0 0.1 0.2 0.3 0.4 0.5
0123456
library normalised expression
frequency(logscale)
18. JACC the Ripper
tailored specifically for scRNA-seq
data
0.0 0.1 0.2 0.3 0.4 0.5
0123456
library normalised expression
frequency(logscale)
incorporates a βtop-downβ
hierarchical classification system
19. JACC the Ripper
tailored specifically for scRNA-seq
data
0.0 0.1 0.2 0.3 0.4 0.5
0123456
library normalised expression
frequency(logscale)
incorporates a βtop-downβ
hierarchical classification system
compiles a report providing
evidence to the user
20. JACC the Ripper
tailored specifically for scRNA-seq
data
0.0 0.1 0.2 0.3 0.4 0.5
0123456
library normalised expression
frequency(logscale)
incorporates a βtop-downβ
hierarchical classification system
compiles a report providing
evidence to the user
21. Three types of bimodality profiles
within scRNA-seq data
22. 1. strong marker genes with many positive cells
Pancreas (mouse, unpublished data)
Testis (human, public data)
38. 3. weak marker genes with many positive cells
0.0000 0.0010 0.0020
02468
library normalised expression
frequency(logscale)
Hhex
0.000 0.002 0.004
02468
library normalised expression
frequency(logscale)
VWF
all other genes
weak marker gene assessment
positive and negative cells
Pancreas (mouse, unpublished data)
Testis (human, public data)
0 10 20 30 40 50 60 70
02468
0 10 20 30 40 50 60 70
02468
UMI counts
frequency(logscale)
Gpx3
observed
expected
heavy tail,
deviates from binomial
0 10 20 30 40 50 60 70
02468
0 10 20 30 40 50 60 70
02468
UMI counts
frequency(logscale)
Canx
observed
expected
no heavy tail,
agrees with binomial
39. 3. weak marker genes with many positive cells
0.0000 0.0010 0.0020
02468
library normalised expression
frequency(logscale)
Hhex
0.000 0.002 0.004
02468
library normalised expression
frequency(logscale)
VWF
all other genes
weak marker gene assessment
positive and negative cells
grouping of βgene friendsβ
Pancreas (mouse, unpublished data)
Testis (human, public data)
0 10 20 30 40 50 60 70
02468
0 10 20 30 40 50 60 70
02468
UMI counts
frequency(logscale)
Gpx3
observed
expected
heavy tail,
deviates from binomial
0 10 20 30 40 50 60 70
02468
0 10 20 30 40 50 60 70
02468
UMI counts
frequency(logscale)
Canx
observed
expected
no heavy tail,
agrees with binomial
40. JACC the Ripper
tailored specifically for scRNA-seq
data
0.0 0.1 0.2 0.3 0.4 0.5
0123456
library normalised expression
frequency(logscale)
incorporates a βtop-downβ
hierarchical classification system
compiles a report providing
evidence to the user
51. Recursive approach of JACC
dataset
cell type 1 cell type 2
cell type 3 cell type 7
cell type 4
cell type 5 cell type 6
52. JACC the Ripper
tailored specifically for scRNA-seq
data
0.0 0.1 0.2 0.3 0.4 0.5
0123456
library normalised expression
frequency(logscale)
incorporates a βtop-downβ
hierarchical classification system
compiles a report providing
evidence to the user
62. Summary
β’ JACC β novel method tailored specifically for scRNA-seq
data
β’ advantages over standard work-flows:
o identifies rare cell populations
o simple in use - no lengthy parameter optimisation steps
o produces a detailed report β transparency to the user
o keeps user close to true nature of data
β’ impact: could be used in any scRNA-seq research
environment β¨
β’ To appear here: β¨
http://wsbc.warwick.ac.uk/wsbcToolsWebpage/β¨
β’ Developing interactive version of JACC the Ripper