Protein-protein interaction network evaluation for identifying potential drug targets

Fereydoun Hormozdiari, Raheleh Salari, Vineet Bafna, S. Cenk Sahinalp

Research output: Contribution to journalArticle

26 Citations (Scopus)

Abstract

As pathogens evolve effective schemes to overcome the effect of antibiotics, the prevalent "one drug and one drug target" approach is falling behind. We propose novel strategies for identifying potential multiple-drug targets in pathogenic protein-protein interaction (PPI) networks with the goal of disrupting known pathways/complexes. Given a set S of pathogenic pathways/complexes, we first consider computing the minimum number of proteins (with no human orthologs) whose removal from the PPI network disrupts all pathways/complexes. Unfortunately, even the best approximation algorithms for this (NP-hard) problem return too many targets to be practical. Thus, we focus on computing the optimal tradeoff (i.e., maximum ratio) between the number of disrupted essential pathways/complexes and the protein targets. For this "sparsest cut" problem, we describe two polynomial time algorithms with respective approximation factors of |S| and (n: number of nodes). On the Escherichia coli PPI network with nine essential (signaling) paths from the KEGG database, our algorithms show how to disrupt three of them by targeting only three proteins (two of them essential proteins). We also consider the case where there are no available essential pathways/complexes to guide us. In order to maximize the number of disrupted "potential" pathways/complexes, we show how to compute the smallest set of proteins whose removal partitions the PPI network into two almost-equal sized subnetworks so as to maximize the number of potential pathways/complexes disrupted. This approach yields 28 potential targets (four of them known drug targets) on the E. coli PPI network whose removal partitions it to two subnetworks with relative sizes of 1-5.

Original languageEnglish (US)
Pages (from-to)669-684
Number of pages16
JournalJournal of Computational Biology
Volume17
Issue number5
DOIs
StatePublished - May 1 2010
Externally publishedYes

Fingerprint

Protein Interaction Maps
Protein Interaction Networks
Protein-protein Interaction
Pathway
Drugs
Proteins
Target
Evaluation
Protein
Pharmaceutical Preparations
Escherichia coli Proteins
Escherichia Coli
Maximise
Partition
Accidental Falls
Computing
Antibiotics
NP-hard Problems
Best Approximation
Polynomial-time Algorithm

Keywords

  • Algorithms
  • Biochemical networks
  • Computational molecular biology

ASJC Scopus subject areas

  • Modeling and Simulation
  • Molecular Biology
  • Genetics
  • Computational Mathematics
  • Computational Theory and Mathematics

Cite this

Protein-protein interaction network evaluation for identifying potential drug targets. / Hormozdiari, Fereydoun; Salari, Raheleh; Bafna, Vineet; Sahinalp, S. Cenk.

In: Journal of Computational Biology, Vol. 17, No. 5, 01.05.2010, p. 669-684.

Research output: Contribution to journalArticle

Hormozdiari, Fereydoun ; Salari, Raheleh ; Bafna, Vineet ; Sahinalp, S. Cenk. / Protein-protein interaction network evaluation for identifying potential drug targets. In: Journal of Computational Biology. 2010 ; Vol. 17, No. 5. pp. 669-684.
@article{5a1f150f5ccf4ed1827c37533b6881a8,
title = "Protein-protein interaction network evaluation for identifying potential drug targets",
abstract = "As pathogens evolve effective schemes to overcome the effect of antibiotics, the prevalent {"}one drug and one drug target{"} approach is falling behind. We propose novel strategies for identifying potential multiple-drug targets in pathogenic protein-protein interaction (PPI) networks with the goal of disrupting known pathways/complexes. Given a set S of pathogenic pathways/complexes, we first consider computing the minimum number of proteins (with no human orthologs) whose removal from the PPI network disrupts all pathways/complexes. Unfortunately, even the best approximation algorithms for this (NP-hard) problem return too many targets to be practical. Thus, we focus on computing the optimal tradeoff (i.e., maximum ratio) between the number of disrupted essential pathways/complexes and the protein targets. For this {"}sparsest cut{"} problem, we describe two polynomial time algorithms with respective approximation factors of |S| and (n: number of nodes). On the Escherichia coli PPI network with nine essential (signaling) paths from the KEGG database, our algorithms show how to disrupt three of them by targeting only three proteins (two of them essential proteins). We also consider the case where there are no available essential pathways/complexes to guide us. In order to maximize the number of disrupted {"}potential{"} pathways/complexes, we show how to compute the smallest set of proteins whose removal partitions the PPI network into two almost-equal sized subnetworks so as to maximize the number of potential pathways/complexes disrupted. This approach yields 28 potential targets (four of them known drug targets) on the E. coli PPI network whose removal partitions it to two subnetworks with relative sizes of 1-5.",
keywords = "Algorithms, Biochemical networks, Computational molecular biology",
author = "Fereydoun Hormozdiari and Raheleh Salari and Vineet Bafna and Sahinalp, {S. Cenk}",
year = "2010",
month = "5",
day = "1",
doi = "10.1089/cmb.2009.0032",
language = "English (US)",
volume = "17",
pages = "669--684",
journal = "Journal of Computational Biology",
issn = "1066-5277",
publisher = "Mary Ann Liebert Inc.",
number = "5",

}

TY - JOUR

T1 - Protein-protein interaction network evaluation for identifying potential drug targets

AU - Hormozdiari, Fereydoun

AU - Salari, Raheleh

AU - Bafna, Vineet

AU - Sahinalp, S. Cenk

PY - 2010/5/1

Y1 - 2010/5/1

N2 - As pathogens evolve effective schemes to overcome the effect of antibiotics, the prevalent "one drug and one drug target" approach is falling behind. We propose novel strategies for identifying potential multiple-drug targets in pathogenic protein-protein interaction (PPI) networks with the goal of disrupting known pathways/complexes. Given a set S of pathogenic pathways/complexes, we first consider computing the minimum number of proteins (with no human orthologs) whose removal from the PPI network disrupts all pathways/complexes. Unfortunately, even the best approximation algorithms for this (NP-hard) problem return too many targets to be practical. Thus, we focus on computing the optimal tradeoff (i.e., maximum ratio) between the number of disrupted essential pathways/complexes and the protein targets. For this "sparsest cut" problem, we describe two polynomial time algorithms with respective approximation factors of |S| and (n: number of nodes). On the Escherichia coli PPI network with nine essential (signaling) paths from the KEGG database, our algorithms show how to disrupt three of them by targeting only three proteins (two of them essential proteins). We also consider the case where there are no available essential pathways/complexes to guide us. In order to maximize the number of disrupted "potential" pathways/complexes, we show how to compute the smallest set of proteins whose removal partitions the PPI network into two almost-equal sized subnetworks so as to maximize the number of potential pathways/complexes disrupted. This approach yields 28 potential targets (four of them known drug targets) on the E. coli PPI network whose removal partitions it to two subnetworks with relative sizes of 1-5.

AB - As pathogens evolve effective schemes to overcome the effect of antibiotics, the prevalent "one drug and one drug target" approach is falling behind. We propose novel strategies for identifying potential multiple-drug targets in pathogenic protein-protein interaction (PPI) networks with the goal of disrupting known pathways/complexes. Given a set S of pathogenic pathways/complexes, we first consider computing the minimum number of proteins (with no human orthologs) whose removal from the PPI network disrupts all pathways/complexes. Unfortunately, even the best approximation algorithms for this (NP-hard) problem return too many targets to be practical. Thus, we focus on computing the optimal tradeoff (i.e., maximum ratio) between the number of disrupted essential pathways/complexes and the protein targets. For this "sparsest cut" problem, we describe two polynomial time algorithms with respective approximation factors of |S| and (n: number of nodes). On the Escherichia coli PPI network with nine essential (signaling) paths from the KEGG database, our algorithms show how to disrupt three of them by targeting only three proteins (two of them essential proteins). We also consider the case where there are no available essential pathways/complexes to guide us. In order to maximize the number of disrupted "potential" pathways/complexes, we show how to compute the smallest set of proteins whose removal partitions the PPI network into two almost-equal sized subnetworks so as to maximize the number of potential pathways/complexes disrupted. This approach yields 28 potential targets (four of them known drug targets) on the E. coli PPI network whose removal partitions it to two subnetworks with relative sizes of 1-5.

KW - Algorithms

KW - Biochemical networks

KW - Computational molecular biology

UR - http://www.scopus.com/inward/record.url?scp=77953016246&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=77953016246&partnerID=8YFLogxK

U2 - 10.1089/cmb.2009.0032

DO - 10.1089/cmb.2009.0032

M3 - Article

VL - 17

SP - 669

EP - 684

JO - Journal of Computational Biology

JF - Journal of Computational Biology

SN - 1066-5277

IS - 5

ER -