You are on page 1of 24

Oral Glucose Tolerance Test (OGTT) Metabolomics: identification of type 2 diabetes effects on primary metabolism

Dmitry Grapov, et al.

What is type 2 diabetes mellitus (T2DM)? Clinically T2DM is:


type 2 diabetic

Can we do a better job by defining the metabolic signature of T2DM?

National trends for T2DM


1994 2000 2009

No Data

<4.5%

4.5-5.9%

6.0-7.4%

7.5-8.9%

>9.0%

FY2011 Obama administration proposes around $159.3 billion for the Iraq and Afghanistan wars (wikipedia)

Study Design
Cohort Overweight women (12-15, obese sedentary, 100 < glucose < 128 mg/dL )
14 week diet* and exercise intervention Weight management and exercise (4 days/week, 30-40 min)
*same diet for all subjects = same background metabolites due to diet!
Sean Adams

Study Design (cont.)


OGTT / exercise 14 week diet and exercise intervention intervention

Measurements Pre- and post-intervention


During Exercise During oral glucose tolerance test (OGTT)
Primary metabolites (n > 300) by GC/TOF Clinical panel: insulin, glucose, lipids 0, 30, 60, 90, 120 minutes

Study Goals: identify metabolites that are responsive to


exercise, OGTT, and changes in insulin sensitivity

Study factors (X) :


3. OGTT
4. Insulin sensitivity

1. Dietary intervention leading to weight loss and increased fitness 2. Exercise

Questions:

1. how does X affect metabolite baselines and excursion profiles 2. what are the independent metabolic markers for X
* Analyses I will talk about

3. what is the biological consequence of metabolic changes associated with X

Data Properties:
Structure
15 samples x 3210 (48, 150) 321 metabolites x 10 (3210) 2 endpoints 5 time-points (10)

Dimensionality (n x m, 150 x 321)


n m 15 samples 2 endpoints 5 time-points 321 metabolites

Metabolic Properties of Interest: Excursions

Baseline and Area Under the Curve (AUC)

Univariate Statistics:

Intervention associated effects: mixed effects models for changes


in baseline and AUC OGTT or exercise associated effects: one-sample t-Test (AUC 0) or ~2-way ANOVA with repeated measures Some considerations Normality Independence False Discovery Rate

risk = 1-(1-p.value)tests

Getting those significant differences is a function of:


significance level () and power (1- )

effect size (standardized difference in means) sample size (n)

Multivariate Analyses

1. Principal Components Analysis (PCA)


Examples

Unsupervised projection of X based on maximum variance (exploratory)

2. Partial Least Squares Projection to Latent Structures (PLS)


Supervised projection of X based on maximum correlation to Y (test of hypothesis)

Interpreting PCA Results


Variance explained (eigenvalues)

Row (sample) scores and column (variable) loadings

PCA example (OGTT time course data)


glucose

219021

*no scaling or centering

How are scores and loadings related?

Data scaling is very important!


glucose (clinical)

glucose (GC/TOF)

219021

*autoscaling (unit variance and centered)

Intervention adjusted PLS model: Scores


time = 0 120 min.

Loadings on the first latent variable (x-axis) can be used to interpret the multivariate changes in metabolites which are correlated with time during the OGTT

goodness of the model is all about the perspective

Determine in-sample (Q2) and outof-sample error (RMSEP) and compare to a random model permutation tests training/testing

Model Training/Testing
Data set selection/splitting is not trivial
Where should these guys go?

Use networks to interpret statistical and multivariate results within a biochemical and mathematical context
Basic anetwork ingredients Make adjacency matrix (what is connected)
product/precursor (KEGG rpairs) chemical similarity (Tanimoto distances) Dependency (partial correlation)

Vertex and edge attributes


(legend)

Example: OGTT metabolomic network


Vertex Size = importance, |PLS coefficient| Color = direction of change, sign of coeff. Border = pvalue < 0.05 (ANCOVA) Edges Tanimoto similarity > 70

with time: pink () cyan ()

pink () cyan ()

Gaussian Markov Network (intervention)

Gaussian Markov Network (cont.)

Future Goals
ExCytR: Excel + Cytoscape + R = Awesome GUI for generating mapped networks Devium Dynamic Multivariate Data Analysis and Visualization Platform GUI for multivariate analysis of Omics data Successor to imDEV Fork me on GitHub: https://github.com/dgrapov/devium
The stack:

You might also like