Discovering functional interaction patterns in Protein-Protein Interactions Networks

Authors: Mehmet E Turnalp Tolga Can

Presented By: Sandeep Kumar

Background

Availability of genome scale protein network

Understanding topological organization

Identification of conserved subnetworks across

different species

Discover modules of interaction

Predict functions of uncharacterized proteins

Improve the accuracy of currently available networks

Aim of study

Using available functional annotations of proteins in PPI network and look for overrepresented patterns of interactions in the network

Present new frequent pattern identification technique PPISpan

Yeast as a model

Why yeast genomics?

A model eukaryote organism …

Well known PPI network

Saccharomyces cerevisiaeSaccharomyces cerevisiae

PPI Network Protein protein interaction shown by edge

between them indicating physical association in

the form of modification, transport or complex

formation

Interesting conserved interaction patterns

among species

Patterns correspond to specific biological

process

Frequent sub-graphs

A graph (sub graph) is frequent if is support (occurrence frequency) in a given dataset is no less than minimum support

threshold

Example: Frequent Subgraphs

GRAPH DATASET

FREQUENT PATTERNS(MIN SUPPORT IS 2)

)A( )B( )C(

)1( )2(

The Algorithm - PPISpan

Based on gSpan

Modified to adapt for PPI network

Candidate generation

Frequency counting

Algorithm: PPISpan (G, L, minSup)

1. Set the vertex labels in G with GO terms from the desired GO level L

2. S <- all frequent 1-edge graphs in G in frequency based lexicographical order

3. for each edge e in S (in ascending order frequency) do

4. SubGraphs (e, minSup, e)5. Remove e from G

Algorithm: Subpgraphs (s, minSup, ext)

1. If (feasible (s, ext)) 2. If DES code of s != to its minimum DFS code3. return 4. C <- Generate all children of s (by growing an edge,

ext)5. Maximal <- true6. For each c in C (in DFS lexicographical order) do7. If support (c) >= minSup8. Subgraphs (c, minSup, c.ext)9. maximal <- false10. If (maximal)11. output s

Datasets used

1. Database of interacting proteins (DIP) data constructed from high-throughput experiments

1. String Database confidence weighted predicted data

1. WI-PHI weighted yeast interactome enriched for direct physical interactions

Gene Ontology annotations

o Used to assign functional category labels to the proteins in PPI network

o Collaborative effort to address the need of consistent descriptions of the gene products in different databases

o Provides description for biological processes, cellular components, and molecular functions

GO slim terms

Provides a broad overview of the functional categories in GO

GO Slim Molecular Function Terms for S. CerevisiaeTerm ID DefinitionGO:3674 molecular function unknownGO:16787 hydrolase activityGO:16740 transferase activityGO:5515 protein binding…Total of 22 broad functional categories

Research Steps

o Label the nodes with functional categories with GO annotations

o Consider molecular function hierarchy

o Focus on functional interaction patterns in arbitrarily topologies

o Find non-overlapping embeddings using PPISpan

Problems faced

o Noise in PPI networko False positiveso False negativeso Accuracy and specificity of

annotations of proteins

Supporting embedding

o Specific instance of the functional pattern realized by certain proteins in the PPI network

Experiment details

o Implemented in C++o Searched for frequent interaction

patterns of support >= 15

Pattern frequency in different datasets

Number of patterns found

Observation

Most of the patterns are trees

Star topology most abundant

Cycles rare

Comparison with known molecular complexes and pathways

Ignore topology and treat patterns as set of proteins for comparison

Molecular complexes from MIPS (Munich Information Center for Protein Sequences) complex catalogue database

Signaling, transport, and regulatory pathways from KEGG database

Use high quality complexes

cpcount

o Average number of different complexes or pathways the embeddings of a frequent interaction pattern overlaps with

o To speculate on the location of interacting patterns

cpoverlap

o Quantifies the overlap between proteins in an embedding and known complexes and pathways

o Ratio of proteins in an embedding that are members of known functional modules

Observations from comparison

o For some of the observed patterns, topology is more important than underlying functional annotations

o Comparison of all the patterns with random patterns in terms of overlap with MIPS complexes

o Comparison of all the patterns with random patterns in terms of overlap with transport and signaling pathways

Analysis of patterns with MIPS complexes

o Selected patterns from DIP and WI-PHI networks

o Selected patterns from the STRING network

o cpoverlap of selected patterns with respect to MIPS complexes

o cpcount of selected patterns with respect to MIPS complexes

Analysis of patterns with KEGG pathways

o Selected patterns from DIP, STRING and WI-PHI networks

o cpoverlap of selected patterns with respect to transport and signaling pathways

o cpcount of selected patterns with respect to transport and signaling pathways

Some interesting Functional interaction patterns

o A frequent functional interaction pattern in the DIP network

o A frequent functional interaction pattern in the WI-PHI network

o A functional interaction pattern related to the MAPK signaling pathwaysignaling pathways

o A functional interaction pattern related to the SNARE interactions in vesicular transport

Conclusions

o Proposed new frequent pattern identification technique, PPISpan

o utilized molecular function Gene Ontology annotations to assign non-unique labels to proteins of a PPI network

o identified significantly frequent functional interaction patterns

o Frequent patterns offer a new perspective into the modular organization of protein-protein interaction networks

QUESTIONS?

THANK YOU