ICBSN2021_Multivariate Data Analysis and Visualization using MetaboAnalyst.pdf

Multivariate Data Analysis and Visualization using MetaboAnalyst

1. Multivariate Data Analysis and
Visualization using MetaboAnalyst
Zanariah Hashim, PhD
Senior Lecturer,
School of Chemical & Energy Engineering,
Faculty of Engineering, UniversitiTeknologiMalaysia
Head,
Food and Biomaterial Engineering Research Group (FoBERG),
UniversitiTeknologiMalaysia, MALAYSIA
Email:
[email protected]

4. •Statistical analysis of data collected on
more than one response/variable
•Analyzed simultaneously
•The variables may be correlated with each
other
•More complex than univariate analysis
•Basic statistical analysis for data mining
Multivariate Data Analysis (MVA)

5. CampusExperience

6. Types of
MVA
Purpose
of MVA
Data
reduction or
structural
simplification
Sorting and
grouping
Relationship
among
variables
Exploratory Data
Analysis
•Data mining, unsupervised
•To gain insights into large
datasets
•Example: PCA, hierarchical
clustering
Regression Analysis
•Prediction models
•X-y relationship
•Example: PLS
Classification
/Discrimination
Analysis
•Pre-determined classes/
bias
•Example: PLS-DA, OPLS-DA

7. Metabolomics
Biomarker search
Quality evaluation of food
etc…
Sample
Preparation, extraction
of metabolites
Instrument analysis Data Analysis
Rawdata
analysis software
Application
Afterprocessing
an interdisciplinary research among
Bioscience,Analytical Chemistryand Informatics

8. Data analysis workflow
Multivariate data analysis
Data pre-treatment
•Mean-center
•Variance scale
•Auto scale
Instrument analysis
Data pre-processing
Data acquisition
LCMS, GCMS,
NMR, FTIR
•Denoising
•Baseline correct
•Peak detection
•Alignment
•Deconvolution
•Normalization
Data matrix
RT m/z No.1･･･No.m
352.7584 X
11
･･･X
1n
・
・
・
・
・
・
・
・
・
･･･
･･･
･･･
・
・
・
・ ・ x
m1･･･x
mn

9. MetaboAnalyst

10. MetaboAnalyst

11. Example 1 –Principal Component
Analysis (PCA)
Samples: fresh oil and
used oil, analyzed by
FTIR
Approximately 4000 data
points for each sample
(from wavelength 400
cm
-1
to 4000 cm
-1
)

12. Figure (a): Exploratory
analysis using principal
component analysis (PCA) –
separation between fresh oil
and used oil can be obtained
Figure (b) and (c): Loading
values of oil samples at 2922
and 1744 cm
-1
which
differentiate fresh and used
oils
Table: Linear regression
between peak areas at
featured wavelengths
(x) vs. peroxide values
(PV) or conjugated
diene values (CDV) (y)
Example 1 –Principal Component
Analysis (PCA)

13. HCA can
1)Separate between “differential” and
“non-differential” (relative to
control)
2)Determine “clusters” and
similarities within the clusters
Data matrix: 154 yeast strains
(n=3) x 84 metabolites (67 LC-
MS and 17 GC-MS)
Normalized to control (BY4742)
and log
2-transformed
Similarity measurement:
Euclidean distance
Clustering using hierarchical clustering analysis
(HCA)based on metabolic profile similarity
Outer
hierarchy
(more
differential)
Metabolites
Strains
(vs. control)
Lower Higher
Example 2 – Clustering Analysis

14. Example 2 – Clustering Analysis
Lo (vs. control) Hi
Members Increased
metabolites
Decreased
metabolites
Possible regulation
A: ino2∆, ino4∆,
opi1∆, ric1∆
(r= 0.85)
Citrate, trehalose,
succinate, aminoadipic
acid
Proline,guanosine,
bisphosphoglycerate
Inositol and phospholipid regulation
(Ino2,Ino4, Opi1)
(Ambroziak1994,
Graves 2000)
B:mks1∆, rtg3∆,
rtg1∆, rrn10∆
(r= 0.87)
Threonine 2-Oxoglutarate,
aminoadipicacid,
glutamate
Mitochondrialretrograde response
(Rtg1, Rtg3, Mks1)
(Sekito2002, Dilova
2002, Tate 2002)
A
B

15. Example 3 – Discriminant Analysis
(OPLS-DA)
-150
-100
-50
0
50
100
150
-100-80-60-40-200 20406080100
to[1]
t[1]
1
2
SIMCA-P+ 12 - 2011-07-22 16:38:03 (UTC-8)
Ethanol sensitive Ethanol tolerant
S-plot indicates the
contribution of variables
(metabolites) towards
the OPLS-DA modeling
-0.6
-0.4
-0.2
-0.0
0.2
0.4
0.6
-0.30-0.20-0.10-0.000.10 0.200.300.40
p(corr)[1]
p[1]
Glx
Glo
Pyr
Lac
Glr
Fum
Suc
Nic
Mal
Oxo
DHAP
ƒÀ-gl
Shi3PGA
Cit/ iso-c
E4P
Pan
R5P
Ru5P
F6P
G1P
G6P
S7P
RuBP
TMP
CMP
UMP
FBP
AMP
GMP
CDP
UDP
ADP
GDP
FMN
CTP
UTP
ATP
GTP
ADP-Rib
UDP-Glu
ADP-Glu
NAD
NADH
NADP
NADPH
FAD
Acetyl-CoA
Alanine
4-aminobut
Serine
Proline
2-keto-iso
Indole
Valine
Threonine/
Citraconic
Hydroxypro
N-acetyl-L
Leucine/ I
Asparagine
Hydroxyiso
Ornithine
Aspartate
Homocystei
p-aminoben
Glutamate
2-hydroxy-Methionine
Histidine
Orotate
Aminoadipi
Phenylpyru
Phenyllact
Phosphoenosn-glycero
Aconitate
Arginine
2-isopropy
Tyrosine
3-phosphos
3-phosphog
N-acetyl-g
Acetyllysi
N-acetyl-g
2-dehydro-
D-gluconat
Tryptophan
Xanthureni
Cysthathio
Cytidine
Uridine
D-glucosam
Thiamine
2,3-diphos
Inosine
6-phospho-
N-acetyl-D
Glutathion
dTMP
3,5-cyclic
Trehalose/
Xanthosine
S-adenosyl
5-phosphor
dTDP
Trehalose-
CDP-ethano
dCTP
UDP-N-acet
Glutathion
Coenzyme A
SI MC A-P+ 12 - 2011-07-22 15: 16: 39 (U TC -8)
Discriminant analysis can
Find the variables that are responsible
for the separation of two groups
Orthogonal Projections to
Latent Structures Discriminant
Analysis (OPLS- DA)
Prior information (class)
provided to the dataset

16. Summary
•Multivariate data analysis (MVA) is an approach
to analyze complex, multi dimensional data
composed of multiple variables. Examples are
principal component analysis, hierarchical
clustering and discriminant analysis.
•MVA can help in gaining useful insights into large
datasets and obtain relationships among the
variables
•MetaboAnalystis a convenient web-based
platform to perform MVA

17. Email: [email protected]