Metabolic Pathway Signatures in Human Gut Microbiome: Lean vs. Obese Individuals, Study notes of Computer Science

Download Metabolic Pathway Signatures in Human Gut Microbiome: Lean vs. Obese Individuals and more Study notes Computer Science in PDF only on Docsity!

Comparative Analysis of Metabolic

Pathways in Metagenomics

Bo Liu, Adam Bazinet, Marco Adelfio

CMSC858L Spring 2009

Outline

Background

– Metagenomics, Metabolic Pathway

Data Set and Processing

– Problem Statement, 6 Lean/Obese Twins,, Functional Annotation

Previous Approach

Our Methods

1. Distance Calculation

2. Discriminative Power Analysis

3. Greedy Search Pathway Signature

4. Pathway Motifs

Future Work

Metabolic Pathway

• Metabolic Pathway : a series of biochemical

reactions that occur inside a cell

• Edges : catalytic enzymes, proteins encoded by genes

• Nodes : molecular substrates, such as dietary

minerals and vitamins

• KEGG : most widely used database

Problem Statement

• Find functional signatures of lean and obese people in

the human gut microbiome

• Signatures could be differentially abundant organisms,

metabolic pathways, or sets of genes

Turnbaugh et al 2009

Pathway Annotation

An example KEGG pathway

Weighted Pathway

10

Previous Approach

• Comparisons based on sum of counts in each pathway

(Turnbaugh et al , Nature 2009; Dinsdale et al , Nature 2008)

Z-score

z = ( x - μ ) / σ

z is the normalized score

x is the original score

μ is the population mean

σ is the population standard deviation

aka standard score or normal score

Z-score

lean samples / obese samples

raw counts of gene X:

Z-normalized counts of gene X:

t -test example

lean / obese

raw counts of gene X:

t -score: 2.

probability: 0.

null hypothesis: not rejected

(with 95% confidence)

Distance Calculation

Would like a metric that gives the distance

between two individual samples

Existing network alignment techniques are

inapplicable because metabolic network

topologies are fixed

Simple metric: sum up the differences in edge

weights between two individual samples

Normalize by total number of ‘reads mapped’

for a given sample

Distance Calculation (naive)

For each pair of samples, sum the difference

in normalized scores across all genes

Produces a distance matrix

LEAN1 000 4285 5792 5023 3970 3885 5215 4461 4478 5818 4586 5459 LEAN2 4285 000 4278 5391 3812 3813 4134 3895 4767 5332 4836 5430 LEAN3 5792 4278 000 5692 4762 4981 2986 4483 5874 6262 5715 6157 LEAN4 5023 5391 5692 000 3968 3982 5104 4466 3874 4728 3338 4194 LEAN5 3970 3812 4762 3968 000 2726 4460 2915 3764 4623 3606 4359 LEAN6 3885 3813 4981 3982 2726 000 4529 3231 3510 4281 3579 3989 OBESE1 5215 4134 2986 5104 4460 4529 000 3987 5339 5936 5336 5869 OBESE2 4461 3895 4483 4466 2915 3231 3987 000 4258 4951 4215 4907 OBESE3 4478 4767 5874 3874 3764 3510 5339 4258 000 4192 3413 4134 OBESE4 5818 5332 6262 4728 4623 4281 5936 4951 4192 000 4009 3497 OBESE5 4586 4836 5715 3338 3606 3579 5336 4215 3413 4009 000 3246 OBESE6 5459 5430 6157 4194 4359 3989 5869 4907 4134 3497 3246 000

Distance Calculation (naive)

gsi for lean (6 individuals) is: 0. P-value based on permutation test (10000 reps) is: 0. gsi for obese (6 individuals) is: 0 P-value based on permutation test (10000 reps) is: 1