Analysis of Gene Expression Data _______________________ Jhoirene B. Clemente Algorithms and Complexity Lab University of the Philippines Diliman
Analysis of Gene Expression Data
_______________________
Jhoirene B. ClementeAlgorithms and Complexity Lab
University of the Philippines Diliman
Overview
● Definitions● Clustering of Gene Expression Data● Visualizations of Gene Expression Data
Definitions
Gene
Basic unit of heredity in a living organism. It is normally a stretch of DNA that codes for a type of protein or for an RNA chain that has a function in the organism.
Gene Expression Data
Expression level of genes in an individual that is measured through Microarray Technology.
Definitions
Definitions
Definitions
Gene Expression Data
Gene Gene Expression
a
b
c
...
n
Definitions
Gene Expression Data
Gene Gene Expression
a
b
c
...
n
1 Sample
n Samples
Definitions
Gene Sample 1
Sample 1
..... Sample m
a
b
c
...
n
m Samples
n Samples
(n x m) Data Matrix
Definitions
Gene Sample 1
Sample 1
..... Sample m
a
b
c
...
n
m Samples
n Samples
(n x m) Data Matrix
Clustering
Clustering is the unsupervised classification of patterns including observations, data sets and feature vectors into groups called clusters, such that objects in the same cluster are similar to each other while objects in different clusters are dissimilar as possible.
image source:ima.umn.edu
Clustering
Clustering is the unsupervised classification of patterns including observations, data sets and feature vectors into groups called clusters, such that objects in the same cluster are similar to each other while objects in different clusters are dissimilar as possible.
image source:ima.umn.edu
Cluster Analysis
Preprocessing● Filtering● Normalization
Clustering
Analysis
Clustering
Partitional
● K-means Algorithm● X-means Algorithm
Hierarchical
Clustering
Given the (n x m) data matrix, we can
● Cluster the set of genes● Cluster the set of samples● Cluster the set of genes and samples
simultaneously.
Data Set
Data set is a time series gene expression data from a synchronized population of yeast.
Data Set
Data set is a time series gene expression data from a synchronized population of yeast.
Preprocessing
Filtering● Removed genes not involved in cell cycle
regulation● Removed genes belonging to more than one
group
Normalization● All gene expression values range from -1.0 to
1.0.
Data Set
Data matrix (384 genes and 17 samples) with 5 classifications.Groupings based from cell cycle phase activation.
Data Set
Group 1: Resting Phase
Data Set
Group 2: First Growth Phase
Data Set
Group 3: Synthesis Phase
Data Set
Group 4: Second Growth Phase
Data Set
Group 5: Cell Division
Clustering of genes
K-means Algorithm
Given n data points in Rd
1. Assign k initial centers of the k clusters
2. Assign all the data points to the nearest cluster
(Euclidean distance, Manhattan distance, etc.)
3. Adjust the k centers
4. Repeat steps 2 and 3 until convergence
Clustering of genes
K-means Algorithm
Given n data points in Rd
1. Assign k initial centers of the k clusters
2. Assign all the data points to the nearest cluster
(Euclidean distance, Manhattan distance, etc.)
3. Adjust the k centers
4. Repeat steps 2 and 3 until convergencek =5
since we want to approximate the 5 biological classification
Clustering of genes
Initialization
1. Choose the first k centers that will maximize the
distance between the clusters
2. Sort the distances between all the data points
and then choose the k initial points at constant
intervals from the sorted list
3. Use the first k points in the data set as the first k
centers
Clustering of genes
Using k-means clustering, with k =5
Clustering of genes
● Clustering may suggest possible roles for genes with unknown functions
● Clustering the samples or experiments may shed light on new subtypes of diseases.
● Identify which type of treatment is suited for a specific type of cancer.
● Building genetic networks
visualization
Vector FusionNon-metric Multidimensional Scaling (nMDS)Principal Components Analysis (PCA)
Vector fusion
Visualization technique that uses the Single point broken line parallel algorithm
nMDS visualization
Input (Dissimilarity Matrix=|ij|) actual distance● In nMDS, only the rank order of entries is
assumed to contain the significant information.● Thus, the purpose of the non-metric MDS
algorithm is to find a configuration of points whose distances reflect as closely as possible the rank order of the data.
● The transformation is by using a non parametric function f. (monotone regression)
dij= f(d
ij) pseudo-distance
PCA
vector fusion visualization
nmds visualization
nmds visualization
nmds visualization
nmds visualization
nmds visualization
nmds visualization
nmds visualization
References2010: "Non-Metric Multidimensional Scaling and Vector Fusion Visualization of Cell Cycle Independent Gene Expressions for Gene Function Analysis", Clemente J., Salido J.A., (2010), Published in the conference proceedings of National Conference on Information Technology for Education(NCITE) 2010 and Philippine IT Journal Feb 2011 Issue.
2010: "Cluster Analysis for Identifying Genes Highly Correlated with a Phenotype", Clemente J., Undergraduate thesis, Department of Computer Science, University of the Philippines Diliman
Thank you for Listening