computational characterization of biomolecular networks in physiology and disease
DESCRIPTION
Computational characterization of biomolecular networks in physiology and disease. Kakajan Komurov, Ph.D Department of Systems Biology University of Texas MD Anderson Cancer Center. Classical to Systems Biology. Gene 1. Gene 2. . . . Function 1. Function 2. - PowerPoint PPT PresentationTRANSCRIPT
Computational characterization of biomolecular networks in physiology and disease
Kakajan Komurov, Ph.DDepartment of Systems Biology
University of Texas MD Anderson Cancer Center
Classical to Systems Biology
Gene 1
Function 1
Gene 2
Function 2 . . .
Gene/protein/molecule-centric research
Classical to Systems Biology
Phenotype 1
Phenotype 2Phenotype 3 . . .
Classical to Systems Biology
Phenotype 1
Phenotype 2Phenotype 3 . . .
• Systems-level analyses
• High throughput experiments – high content data
• Genomics, proteomics, metabolomis, … - “omics” fields
• Extensive use of computational tools
• Computational systems biology
Computational systems biology
• Studying organizational principles of biological systems– Dynamic structure – function
relationship in biological networks
• Developing computational tools to analyze/interpret large-scale data
Computational systems biology
• Studying organizational principles of biological systems– Dynamic structure – function
relationship in biological networks
• Developing computational tools to analyze/interpret large-scale data
Dynamics of protein interaction networks
Stimulus
Protein network
Gene expression program
Dynamics of protein interaction networks
Stimulus
Protein network
Gene expression program
Remodeling of the network
Dynamic organizational principles in protein networks
Komurov and White (2007), Komurov, Gunes, White (2009)
Dynamic organizational principles in protein networks
Komurov and White (2007), Komurov, Gunes, White (2009)
Cancer systems biology
• Extensive data collection at the whole-genome level– The Cancer Genome Atlas Project– Expression Oncology project– Alliance for Signaling project
• System-level understanding of cellular processes activated in cancer
• Computational methods to maximize analytic power, generate testable hypotheses
Biological complexity
• ~22,000 annotated human genes in RefSeq• ~60,000 known protein-protein interactions in human• Millions of indirect relationships between genes• Typical genomic experiment: millions of data points
Objectives• Analyze data within the context of a priori
information– Physical interactions– Function similarity– Sequence similarity– Co-localization
• Extract most relevant genes/subnetworks– Genes with high data values– Coordinately regulated genes with similar functions– Genes with partially redundant functions
• How to score importance/relevance of a gene/subnetwork to the given experimental context?
NetWalk
• Principle: relevance of a gene depends on its measured experimental value and its connections to other relevant genes
• Random walk – based method for scoring network interactions for their relevance to the supplied data
• Simultaneously assesses the local network connectivity and the data values of genes
• No data cutoffs, assesses the whole data distribution
Transition probability
Deriving node relevance scores
Relevance score at step k
Left eigenvector of the transitionprobability matrix
Deriving Edge Flux (EF) value
Node relevance score = visitation probability
Deriving Edge Flux (EF) value
Edge Flux
Node relevance score = visitation probability
Too much bias towards network topology
Deriving Edge Flux (EF) value
Edge Flux
Normalized Edge Flux
Node relevance score = visitation probability
Background node visitation score
Low dose vs. high dose DNA damage
Statistical analyses using EF values instead of gene valuesIdentifying link communities instead of gene communities
Development of drug resistance in breast cancer
• Lapatinib: drug that blocks activity of HER2 oncoprotein
• Patients with activated HER2 have good initial response to the drug, but develop resistance in a short time
• Our strategy: identify networks supporting the drug resistance of breast cancer cells to lapatinib
Cell culture model of drug resistance in breast cancer
SKBR3 SKBR3-R
SKBR3 SKBR3-R +Lapatinib (1uM)
Perform NetWalk analysis of gene expression datato identify most active networks in lapatinib resistance
Strategy
Over-represented networks in lapatinib resistance
0 0.1 0.5 1 20
0.2
0.4
0.6
0.8
1
1.2
ControlGCGR inhibitor (5uM)
Lapatinib concentration (uM)
Surv
ivin
g fr
actio
n
Drug resistance can be reversed by diabetes drugs
0 0.15625 0.3125 0.625 1.25 2.5 5 100
0.2
0.4
0.6
0.8
1
1.2
SKBR3SKBR3-R
Metformin concentration (mM)
Surv
ival
Acknowledgments• Ph.D Mentor: Michael White, Ph.D• Current Mentor: Prahlad Ram, Ph.D• Ram lab:
– Melissa Muller, Ph.D– Jen-Te Tseng– Sergio Iadevaia, Ph.D
• Ju-Seog Lee, Ph.D• Yun-Yong Park, Ph.D
• Collaborators:– Luay Nakhleh, Ph.D (Rice
University)– Michael Davies, M.D Ph.D (MDA)– Mehmet Gunes, Ph.D (UNR)