0% found this document useful (0 votes)

8 views54 pages

Gene Expression Analysis Techniques

The document discusses gene expression analysis, focusing on differential expression methods such as Fold Change and t-Test, which help identify genes that behave differently in normal versus tumor samples. It emphasizes the importance of understanding these differences for better disease therapy and early disease recognition. Additionally, it covers clustering techniques for organizing gene expression profiles and the significance of multiple testing corrections in gene analysis.

Uploaded by

mariammekawi

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

8 views54 pages

Gene Expression Analysis Techniques

Uploaded by

mariammekawi

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

Gene expression analysis

Ulf Leser: Bioinformatics, Wintersemester 2010/2011 1

Agenda

● Differential expression
●
Fold Change
● t-Test
● Clustering
● Databases

3
Differential Expression - Motivation

Why find genes that behave differently in two classes (e.g. normal and
tumor)?

Better understanding of the genetic circumstances that cause the

difference (disease) hopefully leads to better therapy.

Detection of marker-genes enables the early recognition of diseases

as well as the recognition of subtypes of diseases.

Once a cause is identified, a therapy can become more specific,

more effective and reduce side-effects.

4
Differential Expression

Sample
We have:

N ,...,N : normal samples

1 m
T ,...,T : tumor samples
1 n

We look for: genes with significant differences

between N and T

Compare values of gene X from group N

with those of group T
N = {n1,...,n m}
T = {t ,...,t }
1 n

many methods, here:

Fold change
t-test
Gene

5
What to look for - Scatterplot

one point = one gene

B
A A A
Identical (Probably) outlier:
Background- interesting
Noise genes
A : Gene expression Condition A

B : Gene expression Condition B

6
This lecture

● Differential expression
●
Fold Change
● T-Test
● Clustering
● Databases

7
Fold Change

Fold Change (FC)

mean (T )
FC =log2 ( )=log2 (mean (T ))−log2 (mean (N))
mean (N ) 2

|FC| < 1 not interesting

|FC| > 2 very interesting

Log2

mean(tumor) mean(normal) mean(t) / FC

mean(n)

gene a 16 1 16 4

gene b 0.0625 1 0.0625 -4

gene c 10 10 1 0
gene d 200 1 200 7.65

8
Fold Change– Advantages / Disadvantages

+ intuitive measure
- independent of scatter
Exp Exp

S
- independent of absolut values

Exp Exp

2-fold

→ score based only on the mean of the groups not optimal, include variance!
9
Agenda

● Differential expression
●
Fold Change
● T-Test
● Clustering
● Databases

1
T-test – Hypothesis testing

Hypothesis
H0 Null hypothesis (the one we want to reject)
H1 Alternative hypothesis (logical opposite of H0)

Test statistic
Function of the sample that summarizes the characteristics of the latter
into one number with a known distribution.

Significance level
Probability for a false positive outcome of the test,
the error of rejecting a null hypothesis when it is actually true

P-Value
Probability of obtaining the observed test-statistic or higher under
the assumption, that the null hypothesis holds.

1
Hypothesis Testing
• T-Test (unpaired two-sample)
compares the mean of two unpaired samples
•
Assumption:
•
values normally distributed
Hypothesis:
•

H0 (Null hypothesis): µ1 = µ2 vs. µ1 != µ2 (means are not equal)

• Test statistic: Function of the sample that summarizes the data set into
one value that can be used for hypothesis testing

The greater | t |, the greater the

difference between the means
s12 s22
t = ( y1 - y2 ) +
n1 n2 Ways to get a larger t:
• Bigger difference in means
• Smaller standard deviation

17
Hypothesis Testing – t-Test (Welch Test)

• P-value:
•
Probability of observing your data under the assumption that H is true
0
•
Probability that you will be in error if rejecting H
0
• Significance level (a): Probability of a false positive outcome of the test,
the error of rejecting H 0 when it is actually true

t-distribution a = 0.05, 16 samples

If |t| > |T| we reject H

→ p-value is
a/2 a/2 significant
(p-value < a)

-T T
critical values (look-up table)
18
p-value = The chance of rejecting the null
hypothesis by coincidence
----------------------------
For gene expression analysis we can say: The
chance that a gene is categorized as differential
expressed by mistake
Two-Sample Example

A random sample of 15 students is selected from a

specialized high school and a separate, independent
random sample of 11 students is selected from a
diverse high school.

The question is whether average test scores

are, in general, different between the two high
schools.
H0: µ1 = µ2 versus Ha: µ1 ¹ µ2

20
Two populations:
All students in specialized high school
All students in diverse high school
Two sets of population parameters
µ1, s1 for test scores of students in specialized high school
µ2, s2 for test scores of students in diverse high school
Two samples:
15 students in specialized high school
11 students in diverse high school
Two sets of sample summary statistics
y1 , s1 for test scores of 15 students in specialized high
school
y2 , s2 for test scores of 11 students in diverse high school

21
Steps of Hypothesis Testing

• Determine null and alternative hypothesis

• Select a significance level (alpha)

• Take a random sample from the population of interest

• Calculate a test statistic from the sample that

provides information about the null hypothesis

• Decision

23
Two-sample Welch t-test:
Definition:!!
!
H0:!µ1!=!µ2!versus!Ha:!µ1!!!µ2!
!
based!on!independent!random!samples!from!
normal!populations,!when!
!
"1!and!"2!are!completely!unknown,!
!
an!appropriate!test!statistic!is!
s12 s22
t = ( y1 ! y2 ) +
n1 n2 !
A!size!#!test!is!specified!by!the!rejection!rule:!
!
! Ha+ Reject+H0+if+ !
! µ1!!!µ2! |t|!>!tdfs,#/2! !
! ! * !
25
Example :

A random sample of 15 students is selected from a

specialized high school and a separate, independent random
sample of 11 students is selected from a diverse high school.
The question is whether average test scores are, in general,
different between the two high schools.
Test scores for the two samples of high school students are
summarized as: Specialized Diverse
n1 = 15 n2 = 11
y1 = 504.6 y2 =582.3
s1 = 52.5 s2 = 134.5

Use a = 0.10.
Test scores are assumed to be normally distributed.
26
Solution:
The null and alternative hypotheses are
H0: µ1 = µ2 versus Ha: µ1 ¹ µ2
Or
𝐻! : 𝜇" − 𝜇# = 0 𝑣𝑠 𝐻$ : 𝜇" − 𝜇# ≠ 0

Level of the test α = 0.10.

Rejection region:
(s n1 + s n2 )
2 2 2
1 2 (52.52 /15 + 134.52 /11) 2
df S = = = 12.25
( s1 n1 ) + ( s2 n2 ) (52.5 /15) + (134.5 /11)
2 2 2 2 2 2 2 2

n1 - 1 n2 - 1 15 - 1 11 - 1

So dfS = 12. We reject H0 if |t| ≥ t12,0.10/2 = t12,0.05 = 1.782.

-OR- Reject 𝐻0 if the p-value<0.10.

27
Test statistics:

y1 - y2 504.6 - 582.3
t= = = -1.817
2 2 2 2
s s 52.5 134.5
1
+ 2
+
n1 n2 15 11

Assumption check: test scores are normally distributed.

Conclusion: since |t| = 1.817 > 1.782, we reject H0 and

conclude that average test scores in specialized and
diverse high schools are different.

28
Use Critical Value table to obtain bounds on p-value
(largest level α that H0 is not rejected)
𝜶 𝑡!",$/"
0.02 2.681
0.05 2.179
0.10 1.782

So the p-value is between 0.05 and 0.10.

29
Example

30
Example

Example for Gene B

N = {3.58, 4.14, 3.49, 3.37, T = {3.7, 10.9, 10.3, 3.57,

5.29, 5.06, 3.6} 10.5, 8.18, 3.27}

Hypothesis H :m –m =0 H : m – m != 0
0 N T 1 N T

Significance level a = 0.05

𝒎𝒆𝒂𝒏 𝑻 (𝒎𝒆𝒂𝒏 (𝑵)

Test statistic 𝒕 = = −𝟐. 𝟐𝟕
𝒔𝒅(𝑻)𝟐 𝒔𝒅(𝑵)𝟐
𝒎
- 𝒏

P-value =0.06 We cannot reject H , gene B is

0
not significantly differentially expressed!

31
Multiple Testing Correction

Problem: Microarrays has 22k genes, thus an

α=0.05 leads to approximately 22 000 *
0.05 ~ 1100 FPs.

Solution: Multiple testing correction.

False discovery rate (FDR), the expected
proportion of true null hypotheses rejected in the
total number of rejections.(FDR measures the
expected proportion of incorrectly rejected null
hypotheses, i.e. type I errors)

32
Benjamini – Hochberg (FDR)

1. choose a specific α (e.g. α=0.05)

2. rank all m p-values from smallest to largest

3. correct all p-values: BH(p ) = p * m/i

i i=1,...,m i

4. BH (p) = significant if BH(p) ≤ α

Genes p-value rank BH(p) Significant?

(α=0.05)
Gene A 0.00001 1 0.00001*1000/1=0.01 yes

Gene B 0.0004 2 0.0004*1000/2=0.20 no

Gene C 0.01 3 0.01*1000/3=3.3 no

33
This lecture

● Differential expression
●
Fold Change
● T-Test
● Clustering
● Databases

36
Computational Tasks

• differential expression
– which genes have different expression levels across two groups?

• clustering
– which genes seem to be regulated together?
– which treatments/individuals have similar expression profiles?
Clustering Gene Expression Profiles

• given:
– expression profiles for a set of genes or experiments/individuals
/time points (whatever columns represent)

• do:
– organize profiles into clusters
– instances in the same cluster are highly similar to each other
– instances from different clusters have low similarity to each other
Motivation for Clustering
• Exploratory data analysis
– understanding general characteristics of data
– visualizing data

• Generalization
– infer something about an instance (e.g. a gene) based on how
it relates to other instances
Clustering

Ramaswamy
& Golub 2002

41
Overview (Un)Supervised Learning

Clustering
(Unsupervised learning)

hierarchical k-means SOM

42
Hierarchical Clustering of Expression Data
Hierarchical Clustering: A
Dendrogram

height of bar indicates

degree of distance
within cluster
distance scale

leaves represent instances (e.g. genes)

Clustering

●
Goal:
● Partitioning
● Biological interpretation of subtypes (clusters)
●
Requires:
● (useful) similarity measure
●
Advantages:
● Intuitive
● Simple

46
Hierarchical Clustering - Algorithm

1. Distance measure
Euclidean
Manhattan, etc.
dist(xi , x j ) = å (x i,e - x j ,e )2

2. compute similarity matrix S

3. while |S|>1:
4. determine pair (X,Y) with minimal distance
5. compute new value Z = avg (X,Y)
6. delete X and Y in S, insert Z in S
7. compute new distances of Z to all elements in S
8. visualize X and Y as pair

47
Hierarchical Clustering - graphical
ABCDEFG
A A ACEFG a
B A B A
C B. C C.
D C..
D...
(B,D) ®a D E.. (E,F) ®b
E E F...
F E.... F G....
G F..... G a.....
G......
A ACGab A
B A B CGac
C C C
D C.
G.. (A,b) ®c D G. (C,G) ®d
E a... E a..
F b.... F c...
G G

A A A
B ac d B ae B
C a C C
D c. (d,c) ®e D a
e.
(a,e) ®f D
E d.. E E
F F F
G G G
49
Tree-building methods: UPGMA

UPGMA is
unweighted pair group method
using arithmetic mean

1 2

3
4