Genomic Signal Processing Dr. C.Q. Chang Dept. of EEE.

Genomic Signal Processing

Dr. C.Q. Chang

Dept. of EEE

Outline

• Basic Genomics

• Signal Processing for Genomic Sequences

• Signal Processing for Gene Expression

• Resources and Co-operations

• Challenges and Future Work

Basic Genomics

Genome• Every human cell contains 6 feet of double stranded (ds) DNA• This DNA has 3,000,000,000 base pairs representing 50,000-

100,000 genes• This DNA contains our complete genetic code or genome• DNA regulates all cell functions including response to disease,

aging and development• Gene expression pattern: snapshot of DNA in a cell• Gene expression profile: DNA mutation or polymorphism over

time• Genetic pathways: changes in genetic code accompanying

metabolic and functional changes, e.g. disease or aging.

Gene: protein-coding DNA

Protein

mRNA

DNA

transcription

translation

CCTGAGCCAACTATTGATGAA

PEPTIDE

CCUGAGCCAACUAUUGAUGAA

In more detail(color ~state)

Signal Processing for Genomic Sequences

The Data Set

The Problem• Genomic information is digital letters A, T, C and G• Signal processing deals with numerical sequences,

character strings have to be mapped into one or more numerical sequences

• Identification of protein coding regions• Prediction of whether or not a given DNA segment

is a part of a protein coding region• Prediction of the proper reading frame• Comparing to traditional methods, signal processing

methods are much quicker, and can be even more accurate in some cases.

Sequence to signal mapping

1 , 1 , 1 , 1a j t j c j g j

[ ] [ ] [ 1] / 2 [ 2] / 4y n x n x n x n

Signal Analysis

• Spectral analysis (Fourier transform, periodogram)

• Spectrogram

• Wavelet analysis

• HMT: wavelet-based Hidden Markov Tree

• Spectral envelope (using optimal string to numerical value mapping)

Spectral envelope of the BNRF1 gene from the Epstein-Barr virus

(a) 1st section (1000bp), (b) 2nd section (1000bp),

(c) 3rd section (1000bp), (d) 4th section (954bp)

Conjecture: the 4th quarter is actually non-coding

Signal Processing for Gene Expression

Biological Question

Sample preparationMicroarray

Life Cycle

Data Analysis & Modeling

Microarray Reaction

MicroarrayDetection

Taken from Schena & Davis

cDNA clones(probes)

PCR product amplificationpurification

printing

microarray Hybridise target to microarray

mRNA target)

excitation

laser 1laser 2

emission

scanning

analysis

overlay images and normalise

0.1nl/spot

Image Segmentation

• Simple way: fixed circle method• Advanced: fast marching level set segmentation

Advanced Fixed circle

Clustering and filtering methodsPrincipal approaches:• Hierarchical clustering (kdb trees, CART, gene shaving)• K-means clustering• Self organizing (Kohonen) maps• Vector support machines• Gene Filtering via Multiobjective Optimization• Independent Component Analysis (ICA)Validation approaches:• Significance analysis of microarrays (SAM)• Bootstrapping cluster analysis• Leave-one-out cross-validation• Replication (additional gene chip experiments, quantitative PCR)

ICA for B-cell lymphoma data

Data: 96 samples of normal and malignant lymphocytes.

Results: scatter-plotting of 12 independent components

Comparison: close related to results of hierarchical clustering

Resources and Co-operations

Resources: databases on the internet such as

• GeneBank

• ProteinBank

• Some small databases of microarray data

Co-operations in need:

• First hand microarray data

• Biological experiment for validation

Challenges and Future Work• Genomic signal processing opens a new signal

processing frontier• Sequence analysis: symbolic or categorical signal,

classical signal processing methods are not directly applicable

• Increasingly high dimensionality of genetic data sets and the complexity involved call for fast and high throughput implementations of genomic signal processing algorithms

• Future work: spectral analysis of DNA sequence and data clustering of microarray data. Modify classical signal processing methods, and develop new ones.