Page 1
Wellcome Trust Centre for Human Genetics
Mykrobe predictorA desktop application identifying species
and antibiotic resistance from S. aureus
and M. tuberculosis raw genome
sequence data.
Phelim Bradley
[email protected]
Zamin Iqbal [email protected]
Page 2
How can we make clinical genomics a practical
reality?
Mykrobe predictor is clinical
grade software to enable
sequencing data to be used to
identify species and test for drug
susceptibility.
Mykrobe predictor is:
• Quick
• Usable by non informatics-experts
• Scalable
• As accurate as comparable drug susceptibility tests.
Page 3
Mykrobe replaces post culture DST with whole genome sequencing + software analysis.
Mykrobe predictor fits into clinical workflow.
Page 4
Drag and drop demo
Page 5
Drag and drop demo
Page 6
Drag and drop demo
Page 7
Drag and drop demo
Page 8
Drag and drop demo
Page 9
Drag and drop demo
Page 10
Drag and drop demo
Page 11
Drag and drop demo
Page 12
Future advances may decrease required
culture times
Can we also be robust to decreased culture times?
SequencingCulture
Mykrobe
Page 13
Future advances may decrease required
culture times
Can we also be robust to decreased culture times?
SequencingCulture
Mykrobe
Page 14
Mykrobe has been trained and tested on
1000s of samples
• 987 S. aureus and 3529 M. tuberculosis samples (Illumina HiSeq)
S. aureus:
• Resistance predictions for 12 drugs: Penicillin, Erythromycin,
Ciprofloxacin, Methicillin, Fusidic Acid, Tetracycline, Rifampicin,
Gentamycin, Mupirocin and Vancomycin.
• Gold-standard phenotype = a consensus of antibiotic BSAC disc,
Phoenix™ and Etest.
M. tuberculosis:
• Resistance predictions for 10 1st and 2nd line drugs: Isoniazid,
Rifampicin, Ethambutol, Streptomycin, Amoxicillin, Ofloxacin,
Amikacin, Capreomycin, Kanamycin.
• Gold-standard phenotype = proportion method via Löwenstein-
Jensen slope-based testing.
Page 15
Results – toy example
Sample Predict Truth
1 R R
2 R R
3 R R
4 S R
5 S R
6 S R
7 S R
8 S R
9 S R
10 S R
0123456789
10
dru
g
Resis
tan
t S
am
ple
s
VME = Very major error(incorrect S prediction)
(Correct R predictions)
Page 16
0123456789
10
dru
g
Resis
tan
t S
am
ple
s
Results – toy example
Sample Predict Truth
1 R R
2 R R
3 R R
4 R R
5 R R
6 R R
7 R R
8 S R
9 S R
10 S R
VME = Very major error(incorrect S prediction)
TP (Correct R predictions)
Page 17
0123456789
10
dru
g
Resis
tan
t S
am
ple
s
Results – toy example
Sample Predict Truth
1 R R
2 R R
3 R R
4 R R
5 R R
6 R R
7 R R
8 R R
9 R R
10 R R
TP (Correct R predictions)
Page 18
0123456789
10
dru
g
Resis
tan
t S
am
ple
s
Results – toy example
Sample Predict Truth
11 S S
12 S S
13 S S
14 R S
15 R S
16 R S
17 R S
18 R S
19 R S
20 R S
0123456789
10
dru
g
Su
sce
ptible
Sam
ple
s
TN
ME = Major error(incorrect R prediction)
Page 19
0123456789
10
dru
g
Resis
tan
t S
am
ple
s
Results – toy example
Sample Predict Truth
11 S S
12 S S
13 S S
14 S S
15 S S
16 S S
17 S S
18 R S
19 R S
20 R S
TN
ME = Major error(incorrect R prediction)
0123456789
10
dru
g
Su
sce
ptible
Sam
ple
s
Page 20
0123456789
10
dru
g
Su
sce
ptible
Sam
ple
s
0123456789
10
dru
g
Resis
tan
t S
am
ple
s
Results – toy example
Sample Predict Truth
11 S S
12 S S
13 S S
14 S S
15 S S
16 S S
17 S S
18 S S
19 S S
20 S S
TN
Page 21
Mykrobe predictor is as sensitive and specific
as Disc and Phoenix for S. aureus DST .
• We achieved an overall 0.7% VME rate and 0.5% ME
rate across all antibiotics in S. aureus.
VME – missed R
predictions
ME – missed S
predictions
Page 22
Comparison to comparable MTB DST
• Gold standard resistance predictions are slow.
• Faster options:
• PCR based test (e.g. CEPHEID GeneXpert)
• Limited scope (Only RIF predictions)
• Line probe assay (e.g. HAIN MTBDR)
• Static, difficult to update targets.
• Sequencing (e.g. Mykrobe predictor)
• Flexible, updatable panel.
• Single, quantitative test
Page 23
Ability to detect minor populations has
potential clinical benefit.
• Detect coagulase-negative
staphylococci
• Detect atypical
mycobacteria
• Minor calls improve VME
in 2nd line tuberculosis
drugs.
Page 24
Sequence data has additional benefits
Sequencing based solutions will
• Help us better understand within
patient diversity
• Fit in with a global
surveillance system
Page 25
The Future: A sequencer in your USB stick?
Silicon Revolution (Moore’s Law)
Sequencing Technology
Page 26
The Future: A sequencer in your USB stick?
PEN R
ERY R
CIP R
METH R
FUS S
CLIN R
TET S
RIF S
GEN R
MUP S
TRIM R
VAN S
Page 27
Mykrobe is ready for field trips…
Tablet Smart Phone
Raspberry Pi
Model B
- CHEAP!!
Page 28
• Mykrobe predictor is ready for :
• Windows
• Mac
• Linux
• Basespace (submitted)
• Will be tested in 3 UK hospitals in 2015
Mykrobe predictor
Try it out! - you can
download it now.
mykrobe.com
Page 29
Acknowledgments
PI - Zam Iqbal
• Simon Heys, Bill Huang• Claire Gordon, Tim Walker, Tim Peto, Sarah Walker,
Louise Pankhurst, Antonina Votintseva, Derrick Crook• Mariateresa deCesare, David Buck, Gil McVean.• Stefan Niemann, E Grace Smith, Nazir Ismail, Roland Diel,
Shaheed Omar
Page 30
[email protected] [email protected]
@phelimb @ZaminIqbal
Thanks for listening.
http://mykrobe.com
Page 32
Mykrobe replaces post culture DST with whole genome sequencing + software analysis.
Mykrobe predictor fits into clinical workflow.
M. tuberculosis
Page 33
Mykrobe replaces post culture DST with whole genome sequencing + software analysis.
Mykrobe predictor fits into clinical workflow.
S. aureus
Page 34
Future advances may decrease required
culture times
Can we also make it robust to decreased culture times?
S. aureus
Page 35
M. Predictor can accurately determine target
species.
• Speciation was tuned to be as sensitive and specific to
target species.
• Sensitivity and specificity to detect S. Aureus -100%, 98.5%
• Sensitivity and specificity to detect MTBC - 100%,100%
Page 36
Mykrobe and SeqSphere
Page 37
Mykrobe and KvarQ
0
500
1000
ISO
RIF
ET
H
ST
RE
P
MO
X
OF
X
AM
I
CA
P
KA
N
Suscep
tible
Sa
mple
s
Mykrobe
KvarQ
0
100
200
300
ISO
RIF
ET
H
ST
RE
P
MO
X
OF
X
AM
I
CA
P
KA
N
Re
sis
tan
t S
am
ple
s
Mykrobe
KvarQ