31
Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington, IN

Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Embed Size (px)

Citation preview

Page 1: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Improving Gene Function Prediction Using Gene Neighborhoods

Kwangmin Choi

Bioinformatics ProgramSchool of Informatics

Indiana University, Bloomington, IN

Page 2: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Introduction : PLATCOM (A Platform for Computational Comparative Genomics)

PLATCOM is a system for the comparative analysis of multiple genomes.

PLATCOM consists of 3 components: Databases of biological entities

e.g. fna, faa, ptt, gbk… Databases of relationships among entities

e.g. genome-genome, protein-protein pairwise comparison Mining tools over the databases

The web interface of PLATCOM system is located at http://biokdd.informatics.indiana.edu/kwchoi/platcom/

Page 3: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

PLATCOM Web Interface Frontpage of Genome Plot

Page 4: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Background :What is operon ? http://biocyc.org:1555/ECOLI/new-image?object=Transcription-Units

The operon structure was found in 1960 by 2 French biologists. Jacob,F. and Monod,J. (1961) Genetic regulatory mechanisms in the synthesis of proteins. J. Mol. Biol., 3, 318–356.

An operon is a group of genes that encodes functionally linked proteins. Its components are :

Adjacent (200-300 nt) On the same strand (+ or -) Co-expressed by one promoter.

Page 5: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Background :How to identify or predict operon structure?

When a promoter and terminator are known : Gene clusters = Transcription Units Classical concept of operon

When a promoter is not known : Gene clusters = Directrons Hypothetical operon candidates Depending on direction and proper intergenic distance (200-

300 nt)

Computational methods have been developed to find gene clusters in bacterial genomes.

Page 6: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

PCBBH and PCHR.Overbeek et al. PNAS, 1999, Vol.96, pp.2896-2901

PCBBH : Pair of Close Bidirectional Best Hits

BBH : Bidirectional Best Hits

PCH : Pair of Close Homologs

COG : Clusters of Orthologous Genes

Page 7: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Background :Über-operon : P.Bork et al. Treds. Biochem. Sci., Vol. 25, pp. 474-479

Über-operon : A set of genes with a close functional and regulatory contexts that tends to be conserved despite numerous rearrangements.

This concept focus on the functional themes of operons, not a specific genes or gene order.

Page 8: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Background :Why gene clusters are conserved ?

Certain operons, particularly those that encode subunits of multiprotein complexes (e.g. ribosomal proteins) are conserved in phylogenetically distant bacterial genomes.

These gene clusters might have been conserved since the last universal common ancestor. Why?

Selfish-operon hypothesis :Horizontal transfer of an entire operon is favored by natural selection over transfer of individual genes because co-expression and co-regulation are preserved.

Page 9: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Background : Problems in Operon Prediction.

Over 150 genomes have been fully sequenced until today, but The biological functions of some genes are still unknown.

There is only a few promoter detection algorithms, but they are not fully satisfactory.

In many cases, genomic data files do not provide full information of genes and their products. ( e.g. gene name, COG, PID.)

Operon tends to undergo multiple rearrangements during evolution.

As a result, gene order at a lever above is poorly conserved. (e.g. genes involved in de novo purine synthesis)

Page 10: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Background : Problems in Computational Algorithms to Predict Operons

Direct Signal Finding Experiment-based approach Transcription promoters (5’-end) and terminators (3’-end) were

searched. Only be effective for species whose transcription signals are well

known, E.coli.

Combination of gene expression data, functional annotation and other experimental data.

Literature-based approach Primarily applicable to well studied genomes such as E.coli, because

data files are incomplete for other genomes. In many cases, genomic data files do not provide full information of

genes and their products. ( e.g. gene name, COG, PID.)

Page 11: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Procedure

As a part of PLATCOM project, an integrated whole genome analysis system was built on BIOKDD server.

Web interface for all-to-all pairwise comparison DB and tools are also provided.

Several tools for multiple genomes analysis were written in Perl and then gene neighborhoods was reconstructed from the clustering data.

My gene clustering algorithm was used to compensate the defect of the literature-based approach.

Connected gene neighborhoods were analyzed to predict gene function and functional coupling between clusters.

Page 12: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Materials/ Tools

Raw Data 22 genomes were chosen for this study. (14 groups) Protein-Protein Pairwise Comparison Data

e.g. http://biokdd.informatics.indiana.edu/kwchoi/Thesis/L42023.faa.U00096.faa.cmp.txt

PTT files from NCBI site e.g. http://biokdd.informatics.indiana.edu/kwchoi/Thesis/U00096.ptt.txt

Data Generated by Web Tools Gene Clustering Data (based on sequence homology)

e.g. http://biokdd.informatics.indiana.edu/kwchoi/Thesis/clustering_13321_23_750.txt Gene Clusters generated from PTT file (given intergenic distance)

e.g. http://biokdd.informatics.indiana.edu/kwchoi/Thesis/candidates_22211.htm

E. coli database for reality check http://biocyc.org/ http://ecocyc.org/

Page 13: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Genomes http://www.infobiogen.fr/services/deambulum/english/genomes2a.html

Page 14: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Procedure My Approach to reconstruct Genomic Neighborhoods

The idea underlying this study is that Different genomes contain different, overlapping parts of evolutionarily and

functionally connected gene neighborhoods By generating a “Tiling Path”, the entire neighborhood can be

reconstructed.

Genomic context of well-known genome (e.g. E.coli ) is used as a contextual framework.

Start with looking at this framework and then search a group of similar gene neighborhoods in the target genomes.

“Genomic context” means the pattern of series of COG. If COG is not given, we can predict the function of a unknown gene based on my gene clustering data.

We can also identify some “Hitchhikers”. “Hitchhikers” are inserted genes that are originated from different contexts/themes.

Page 15: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Tiling PathV.Koonin et al. Nucleic Acids Research, 2002, Vol.30, No.10, pp. 2212-2223

Page 16: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Gene Neighborhoods

Page 17: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Results Case 1

Relationship between Gene Order and Phylogenetic Distance

Case 2 One theme : Typical Operon (rbs operon)

Reconstruct gene neighborhoods Find missing components from the reconstructed gene clusters.

Case 3 Two or more themes : Functional Coupling ?

Find genomic hitchhikers Predict gene function of uncharacterized protein Predict functional coupling

Page 18: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Case 1 : Gene Order and Phylogenetic Distance

If gene order of two genome is well conserved, the sequence of homologs should appear as a line on the genome comparison diagonal plot.

What is the relationship between phylogenetic distance and the conservation of gene order?

Page 19: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Phylogenetic TreeV.Daubin et al. Genome Research, Vol 12, Issue 7, 1080-1090

Page 20: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Genome Comparison Diagonal Plot

: Phylogenetically-Distant Species (Z-score = over 500)

Page 21: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Genome Comparison Diagonal Plot

: Phylogenetically-Close Species (Z-score > 1000)

Page 22: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Fragmented Gene Clusters

Page 23: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Case 1 : Conclusion

Gene order in phylogenetically-distant species are poorly conserved.

But this observation does not mean that gene order is conserved very well among the phylogenetically-close species.

In case of very close species (e.g. E.coli vs. H.influenza), gene orders are completely scattered.

In most cases, only a small number of genes are observed as a short line or cluster and we may consider it as a putative operon.

In next step, this possibility will be investigated deeply.

Page 24: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Case 2 : Rbs Operon (Typical Operon)

Theme : Ribose transport across membrane COG1869 D-ribose high-affinity transport system; membrane-associated protein COG1129 ATP-binding component of D-ribose high-affinity transport system COG1172 D-ribose high-affinity transport system COG1879 D-ribose periplasmic binding protein COG0524 ribokinase COG1609 regulator for rbs operon

http://biocyc.org:1555/ECOLI/new-image?type=OPERON&object=TU00206

Page 25: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Case 2 : Rbs OperonZ-score = over 750, Intergenic Distance = 300

Page 26: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Case 2 : Conclusion

All components are involved in ribose transport across bacterial cell membrane

In Rbs operon system, gene order pattern is 1869-1129-1172-1879-0524-1609.

10 out of 22 genomes have this operon system. Exceptsome cases, this gene order pattern is conserved very well.

So it is possible that there exists a kind of “General Contextual Framework” of gene order.

Page 27: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Case 3 : Functional Coupling of 2 or more themes

Theme 1 : Transcription COG0779 Uncharacterized Conserved Protein COG0195 Transcription elongation factor COG2740 Predicted nucleic-acid-binding protein (transcription termination?)

Theme 2 : Translation COG1358 Ribosomal protein S17E COG0532 Translation initiation factor 2 (GTPase) COG1550 Uncharacterized Conserved Protein COG0858 Ribosome-binding factor A COG0184 Ribosomal protein S15P/S13E COG0130 tRNA Pseudouridine synthase

Hitchhiker ? COG0196 FDA Synthase (Hitchhiker?)

http://biocyc.org:1555/ECOLI/new-image?type=OPERON&object=TU341

Page 28: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Case 3 : Functional CouplingZ-score = over 750, Intergenic Distance = 300

Page 29: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Case 3 : Conclusion

Functional Coupling : In bacteria, transcription, translation and RNA modification/degradation are

coupled and the advantages of co-regulation the corresponding genes are obvious.

COG0779(Uncharacterized) is almost inseparable from the COG0195(Transcription Elongation Factor), so it is likely to be a functional partner of COG0195.

Hitchhiker : The association of the COG0196(FDA synthase) is not as tight as the

connections between the genes belonging to the theme.

Gene function prediction : The functions of 3 genes in AE0004092 genomes can be predicted by

reading genomic context.

Page 30: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Conclusion Genome Comparison Diagonal Plot visualizes the

sequence comparison of 2 genomes. It is a simple tool, but presents a very strong intuition to understand the genome structure.

Conserved gene neighborhoods reconstructed from many genomes by the Tiling Path Method can be used to predict the functions of uncharacterized genes and functional coupling between well-characterized genes in those genomes.

Ultimately, We can use this methods to reconstruct metabolic and functional subsystems.

Page 31: Improving Gene Function Prediction Using Gene Neighborhoods Kwangmin Choi Bioinformatics Program School of Informatics Indiana University, Bloomington,

Acknowledgements

Haifeng Zhao Genome Pairwise Comparison DB

Scott Martin Server Management and Technical Suppor

Dr. Sun Kim Graduate Advisor and P.I.