Using Networks to Integrate Omic and Semantic Data: Towards Understanding Protein Function on a Genome Scale

Size: px

Start display at page:

Download "Using Networks to Integrate Omic and Semantic Data: Towards Understanding Protein Function on a Genome Scale"

Blaise Miles
6 years ago
Views:

Using Networks to Integrate Omic and Semantic Data: Towards Understanding Protein Function on a Genome Scale Biomarker Data Analysis.10.01, 9:25-9:50 Mark B Gerstein Yale (Comp. Bio. & Bioinformatics) Slides downloadable from Lectures.

1 Using Networks to Integrate Omic and Semantic Data: Towards Understanding Protein Function on a Genome Scale Biomarker Data Analysis.10.01, 9:25-9:50 Mark B Gerstein Yale (Comp. Bio. & Bioinformatics) Slides downloadable from Lectures.GersteinLab.org (Please read permissions statement.) Paper references mostly from Papers.GersteinLab.org. See streams.gerstein.info on photos & images (Very Short Nets talk: predess + SANDY [I:IBIOMARKER]) 1 1 Gerstein.info/talks Mark Gerstein,

The problem: Grappling with Function on a Genome Scale?. ~530 250 of ~530 originally characterized on chr.

2 The problem: Grappling with Function on a Genome Scale?. ~ of ~530 originally characterized on chr. 22 [Dunham et al. Nature (1999)] >25K Proteins in Entire Human Genome (with alt. splicing) 2 2 Gerstein.info/talks Mark Gerstein,

EF2_YEAST Descriptive Name: Elongation Factor 2 Lots of references to papers Summary sentence describing function: This protein promotes the GTP-dependent translocation of the

3 EF2_YEAST Descriptive Name: Elongation Factor 2 Lots of references to papers Summary sentence describing function: This protein promotes the GTP-dependent translocation of the nascent protein chain from the A-site to the P-site of the ribosome. Traditional single molecule way to integrate evidence & describe function 3 3 Gerstein.info/talks Mark Gerstein,

4 Some obvious issues in scaling single molecule definition to a genomic scale Fundamental complexities Often >2 proteins/function Multi-functionality: 2 functions/protein Role Conflation: molecular, cellular, phenotypic 4 4 Gerstein.info/talks Mark Gerstein,

molecular, cellular, phenotypic Fun terms but do they scale?

5 Some obvious issues in scaling single molecule definition to a genomic scale Fundamental complexities Often >2 proteins/function tinman.nikunnakki.info Multi-functionality: 2 functions/protein Role Conflation: molecular, cellular, phenotypic Fun terms but do they scale?... Starry night (P Adler, 94) [Seringhaus et al. GenomeBiology ()] 5 5 Gerstein.info/talks Mark Gerstein,

6 Hierarchies & DAGs of controlled-vocab terms but still have issues... MIPS (Mewes et al.) GO (Ashburner et al.) [Seringhaus & Gerstein, Am. Sci. '08] 6 6 Gerstein.info/talks Mark Gerstein,

Towards Developing Standardized Descriptions of Function Subjecting each gene to standardized expt. and cataloging effect KOs of each gene in a variety of std.

7 Towards Developing Standardized Descriptions of Function Subjecting each gene to standardized expt. and cataloging effect KOs of each gene in a variety of std. conditions => phenotypes Std. binding expts for each gene (e.g. prot. chip) Function as a vector Interaction Vectors [Lan et al, IEEE 90:1848] 7 7 Gerstein.info/talks Mark Gerstein,

8 8 8 Gerstein.info/talks Mark Gerstein, Networks (Old & New) Classical KEGG pathway Same Genes in High-throughput Network [Seringhaus & Gerstein, Am. Sci. '08]

Networks occupy a midway point in terms of level of

9 Networks occupy a midway point in terms of level of understanding 1D: Complete Genetic Partslist [Fleischmann et al., Science, 269 :496] ~2D: Bio-molecular Network Wiring Diagram [Jeong et al. Nature, 41:411] 3D: Detailed structural understanding of cellular machinery 9 9 Gerstein.info/talks Mark Gerstein,

10 Networks as a universal language Internet [Burch & Cheswick] Disease Spread [Krebs] Protein Interactions [Barabasi] Food Web Electronic Circuit Social Network Gerstein.info/talks Mark Gerstein, Neural Network [Cajal]

Combining networks forms an ideal way of integrating diverse information Metabolic pathway Part of the TCA cycle Transcriptional regulatory network

11 Combining networks forms an ideal way of integrating diverse information Metabolic pathway Part of the TCA cycle Transcriptional regulatory network Physical proteinprotein Interaction Co-expression Relationship Genetic interaction (synthetic lethal) Signaling pathways Gerstein.info/talks Mark Gerstein,

12 12 12 Gerstein.info/talks Mark Gerstein, pers. photo, see streams.gerstein.info Standardized Phenotypic Functions for Each Gene: Predicting Essential Genes

13 What are essential genes? Not all genes are necessary Redundancy Condition-specific Function not vital Essential genes are those necessary for survival Deletion / disablement is lethal Attractive drug targets Disable one gene, kill the pathogen Essentiality depends on growth environment Normal definition: Laboratory growth, rich medium 13

14 Identifying essential genes Large-scale experimental screens PCR deletion, transposon disruption, RNAi Model organisms Homology mapping Map findings to similar genes in other species Conservation as essentiality 1 2 E 3 14

15 Estimating essentiality by homology Gene from close relative Saccharomyces mikatae S. mikatae ORFN2098 S. cerevisiae YBL042W ESSENTIAL NON-ESSENTIAL BLAST vs. S.cerevisiae 15

16 The trouble with homology Sequence homology has limitations Some fraction of essential genes are unique to a given organism For drug-design design purposes: Don t t want to attack genes that are broadly conserved! 1 2 E 3 Good target 16

17 Approach 1. Gather attributes to characterize essential genes 2. Develop a computational predictor based on these attributes S. cerevisiae (learn) S. mikatae (predict) 3. Assess predictive success Homology,, permutation, knockouts Seringhaus et al. Gen. Res. 16:1126 ('06) 17

18 Group features into Sequence 3 classes Gene GC content, codon usage, length, etc. Genome Chromosome position, duplication Predicted from Sequence TM helix, localization, predicted hydrophobicity Close-stop Gene length TGC TGT TGG TGA Experimental Data Gene expression, interaction, localization Seringhaus et al. Gen. Res. 16:1126 ('06) 18

19 28 features in 3 classes Gene Seq Genome Seq Predicted from Seq Experimental Data Seringhaus et al. Gen. Res. 16:1126 ('06) NAME DESCRIPTION TYPE CAI codon adaptation index real Nc effective number of codons real GC GC content real L_aa length of putative protein in amino acids integer CLOSE_STOP_RATIO percentage of codons one third-base away from a stop codon real RARE_AA_RATIO percentage rare amino acids in translated ORF real BLAST_yeast # BLAST hits in yeast genome (measure of duplication) integer BLAST_closeyeast hits in # of 6 close yeast species integer BLAST_proks hits in # of 5 prokaryotic species integer CHR located on chromosome (number) integer CHR_POSN Position relative to end of chromosome real INTRON gene has intron? binary Hydro hydrophobicity score real TM_HELIX Number of predicted transmembrane helices (TMHMM 2.0) integer PRED_mitochondria predicted subcellular localization: mitochondria binary PRED_cytoplasm predicted subcellular localization: cytoplasm binary PRED_er predicted subcellular localization: endoplasmic reticulum binary PRED_nucleus predicted subcellular localization: nucleus binary PRED_vacuole predicted subcellular localization: vacuole binary PRED_other predicted subcellular localization: any other compartment binary mrna_expr normalized mrna expression in copies/cell real INTXN_PARTNERS protein-protein interaction partners integer EXPT_mitochondria experimental subcellular localization: mitochondria binary EXPT_cytoplasm experimental subcellular localization: cytoplasm binary EXPT_er experimental subcellular localization: endoplasmic reticulum binary EXPT_nucleus experimental subcellular localization: nucleus binary EXPT_vacuole experimental subcellular localization: vacuole binary EXPT_other experimental subcellular localization: any other compartment binary 19

20 14 features (unstudied genomes) Gene Seq Genome Seq Predicted from Seq Experimental Data Seringhaus et al. Gen. Res. 16:1126 ('06) NAME DESCRIPTION TYPE CAI codon adaptation index real Nc effective number of codons real GC GC content real L_aa length of putative protein in amino acids integer CLOSE_STOP_RATIO percentage of codons one third-base away from a stop codon real RARE_AA_RATIO percentage rare amino acids in translated ORF real BLAST_yeast # BLAST hits in yeast genome (measure of duplication) integer BLAST_closeyeast hits in # of 6 close yeast species integer BLAST_proks hits in # of 5 prokaryotic species integer CHR located on chromosome (number) integer CHR_POSN Position relative to end of chromosome real INTRON gene has intron? binary Hydro hydrophobicity score real TM_HELIX Number of predicted transmembrane helices (TMHMM 2.0) integer PRED_mitochondria predicted subcellular localization: mitochondria binary PRED_cytoplasm predicted subcellular localization: cytoplasm binary PRED_er predicted subcellular localization: endoplasmic reticulum binary PRED_nucleus predicted subcellular localization: nucleus binary PRED_vacuole predicted subcellular localization: vacuole binary PRED_other predicted subcellular localization: any other compartment binary mrna_expr normalized mrna expression in copies/cell real INTXN_PARTNERS protein-protein interaction partners integer EXPT_mitochondria experimental subcellular localization: mitochondria binary EXPT_cytoplasm experimental subcellular localization: cytoplasm binary EXPT_er experimental subcellular localization: endoplasmic reticulum binary EXPT_nucleus experimental subcellular localization: nucleus binary EXPT_vacuole experimental subcellular localization: vacuole binary EXPT_other experimental subcellular localization: any other compartment binary 20

21 Correlating features to essentiality Pair-wise correlational analysis Each feature vs. essentiality (Nomogram( Nomogram) Not looking at co-variation or higher-order effects 3 features have pair-wise correlations that were not significant (p > 0.05), and were set to zero Seringhaus et al. Gen. Res. 16:1126 ('06) 21 Seringhaus et al., Genome Res Sep;16(9):

22 Simple & Complex Classifier Design Simple Naive Bayes (just adding votes of each feature in nomogram) Classifier Complex: hybrid machine learning system Combination of Naïve Bayes,, decision trees, & logistic regression Naive Bayes classifier decision stump boosted using Adaboost (Freund and Schapire 1996) random forest (Breiman( 2001) alternating decision tree with 10 boosting iterations (Freund and d Mason 1999) C4.5 decision tree (Quinlan 1993) multinomial logistic regression model with ridge estimator (le Cessie and van Houwelingen 1992) Seringhaus et al. Gen. Res. 16:1126 ('06) 22

23 ROC curve: classifier performance on S cerevisiae Receiver-Operating Characteristic curve Cost-benefit analysis of a binary classifier as discrimination threshold is varied Ideal predictor: Top left Random predictor: Diagonal Area of interest: Near origin Don t t need to identify every essential gene Adjust cost matrix to discourage false positives Seringhaus et al., Genome Res Sep;16(9):

24 Classifier applied to S. mikatae ORF mitochondria cytoplasm er nucleus vacuole other CAI Nc GC L_aa Hydro CLOSE_STOP_RATIO RARE_AA_RATIO TM_HELIX ESS ORFN: ? ORFN: ? ORFN: ? ORFN: ? ORFN: ? ORFN: ? ORFN: ? ORFN: ? ORFN: ? ORFN: ? ORFN: ? ORFN: ? ORFN: ? ORFN: ? ORF mitochondria cytoplasm er nucleus vacuole other CAI Nc GC L_aa Hydro CLOSE_STOP_RATIO RARE_AA_RATIO TM_HELIX ESS ORFN: ORFN: ORFN: ORFN: ORFN: ORFN: ORFN: ORFN: ORFN: ORFN: ORFN: ORFN: Assigns essentiality score to each gene ORFN: ORFN: Seringhaus et al. Gen. Res. 16:1126 ('06)

25 S.mikatae SCORE RANK ORF_ID (0-1) Assigned essentiality score (0-1) 3939 putative S. mikatae ORFs How does this compare to homology mapping? S. mikatae ORF1 BLAST vs. S.cerevisiae S. cerevisiae homolog ESSENTIAL NON-ESSENTIAL 26 Seringhaus et al., Genome Res Sep;16(9):

26 S.mikatae SCORE homolog in BLAST vs. S.cerevisiae essential genes RANK ORF_ID (0-1) S.cerevisiae 1.00E E E Multiple YOR046C YOR341W YIL126W YDL031W YJL050W YBR088C Multiple YMR308C YDR190C YDR101C YJL080C YLL004W YOL010W Multiple YPL235W YBR247C YOR117W YOR116C YBR158W YBR119W YGL078C YMR290C YGR145W YBL023C YOL021C YKR081C YOR259C YPL228W YPL211W YDR525W-A YNL101W YDL173W Seringhaus et al. Gen. Res. 16:1126 ('06)

S.mikatae SCORE homolog in BLAST vs. S.cerevisiae essential genes SELECTED FOR RANK ORF_ID (0-1) S.cerevisiae 1.00E-04 1.00E-20 1.00E-50 KNOCKOUT?

974 YDL031W 1 1 1 Predicted Essential 6 12016 0.964 YJL050W 1 1 1 Predicted Essential 7 911 0.964 YBR088C 1 1 1 Predicted Essential 8 22659 0.962 Multiple 1 1 1 Predicted Essential 9 17592 0.

27 S.mikatae SCORE homolog in BLAST vs. S.cerevisiae essential genes SELECTED FOR RANK ORF_ID (0-1) S.cerevisiae 1.00E E E-50 KNOCKOUT? Multiple Predicted Essential YOR046C Predicted Essential YOR341W Predicted Essential YIL126W Predicted Essential YDL031W Predicted Essential YJL050W Predicted Essential YBR088C Predicted Essential Multiple Predicted Essential YMR308C Predicted Essential YDR190C Predicted Essential YDR101C Disputed YJL080C Disputed YLL004W Predicted Essential YOL010W Predicted Essential 10 / 10 agree with homology mapping Multiple Predicted Essential YPL235W Predicted Essential YBR247C Predicted Essential YOR117W Predicted Essential YOR116C Predicted Essential YBR158W Disputed YBR119W Disputed YGL078C Disputed YMR290C Predicted Essential YGR145W Predicted Essential YBL023C Predicted Essential YOL021C Predicted Essential YKR081C Predicted Essential YOR259C Predicted Essential YPL228W Predicted Essential YPL211W Predicted Essential YDR525W-A Predicted Non-Essential 25 / 30 agree with homology mapping YNL101W Predicted Non-Essential YDL173W Predicted Non-Essential 28 Seringhaus et al. Gen. Res. 16:1126 ('06)

28 Using Networks to Describe Function Gerstein.info/talks Mark Gerstein, [NY Times, 2-Oct-2005]

29 Dynamic Yeast TF network Transcription Factors Analyzed network as a static entity Target Genes But network is dynamic Different sections of the network are active under different cellular conditions Integrate gene expression data Luscombe et al. Nature 431: Gerstein.info/talks Mark Gerstein,

30 Gene expression data for five cellular conditions in yeast Multi-stage Binary Cellular condition Cell cycle Sporulation Diauxic shift DNA damage Stress response [Brown, Botstein, Davis.] Gerstein.info/talks Mark Gerstein,

31 Backtracking to find active sub-network Define differentially expressed genes Identify TFs that regulate these genes Identify further TFs that regulate these TFs Active regulatory sub-network Gerstein.info/talks Mark Gerstein,

32 Network usage under different conditions Luscombe et al. Nature 431: 308 Mark Gerstein, 3333Gerstein.info/talks static

33 34 34 Gerstein.info/talks Mark Gerstein, Network usage under different conditions cell cycle

34 35 35 Gerstein.info/talks Mark Gerstein, Network usage under different conditions sporulation

35 36 36 Gerstein.info/talks Mark Gerstein, Network usage under different conditions diauxic shift

36 37 37 Gerstein.info/talks Mark Gerstein, Network usage under different conditions DNA damage

37 38 38 Gerstein.info/talks Mark Gerstein, Network usage under different conditions stress response

38 Network usage under different conditions Cell cycle Sporulation Diauxic shift DNA damage Stress SANDY: 1. Standard graph-theoretic statistics: - Global topological measures - Local network motifs 2. Newly derived follow-on statistics: - Hub usage - Interaction rewiring 3. Statistical validation of results Luscombe et al. Nature 431: Gerstein.info/talks Mark Gerstein,

Standard graph-theoretic statistics: - Global topological

Newly derived follow-on statistics: - Hub usage -

39 Network usage under different conditions Cell cycle Sporulation Diauxic shift DNA damage Stress SANDY: 1. Standard graph-theoretic statistics: - Global topological measures - Local network motifs 2. Newly derived follow-on statistics: - Hub usage - Interaction rewiring 3. Statistical validation of results Gerstein.info/talks Mark Gerstein,

40 Outdegree Indegree Pathlength Clustering coefficient Cell cycle Multi-stage Controlled, ticking over of genes at different stages Sporulation Diauxic shift DNA damage Stress response Binary Quick, large-scale turnover of genes Analysis of conditionspecific subnetworks in terms of global topological statistics Luscombe et al. Nature 431: Gerstein.info/talks Mark Gerstein,

41 Singleinput module Multi-input module Feedforward loop 32% 24% 44% Cell cycle 39% 17% 45% Sporulation Multi-stage Controlled, ticking over of genes at different stages 57% 24% 19% Diauxic shift 56% 27% 17% DNA damage Binary Quick, large-scale turnover of genes 59% 20% 21% Stress response Analysis of conditionspecific subnetworks in terms of occurrence of local motifs Luscombe et al. Nature 431: Gerstein.info/talks Mark Gerstein,

Cell cycle Sporulation Diauxic shift DNA damage Stress multi-stage conditions binary conditions Summary less pronounced Hubs more pronounced

42 Cell cycle Sporulation Diauxic shift DNA damage Stress multi-stage conditions binary conditions Summary less pronounced Hubs more pronounced longer Path Lengths shorter more TF inter-regulation less complex loops (FFLs) Motifs simpler (SIMs) Gerstein.info/talks Mark Gerstein,

43 Transient Hubs Regulatory hubs Questions: Do hubs stay the same or do they change over between conditions? Do different TFs become important? Our Expectations Literature: Hubs are permanent features of the network regardless of condition Random networks (sampled from complete regulatory network) Random networks converge on same TFs 76-97% overlap in TFs classified as hubs (ie hubs are permanent) Luscombe et al. Nature 431: Gerstein.info/talks Mark Gerstein,

44 cell cycle sporulation diauxic shift DNA damage stress response all conditions transitient transient hubs permanent hubs Some permanent hubs house-keeping functions Most are transient hubs Different TFs become key regulators in the network Implications for conditiondependent vulnerability of network Luscombe et al. Nature 431: Gerstein.info/talks Mark Gerstein,

45 cell cycle Swi4, Mbp1 Ime1, Ume6 sporulation diauxic shift DNA damage stress response all conditions transitient hubs Msn2, Msn4 permanent hubs Luscombe et al. Nature 431: Gerstein.info/talks Mark Gerstein,

46 cell cycle sporulation diauxic shift DNA damage stress response all conditions transitient hubs permanent hubs Unknown functions Luscombe et al. Nature 431: Gerstein.info/talks Mark Gerstein,

47 Conclusions Developing Standardized Descriptions of Protein Function Integration of Information for Prediction and Description Prediction of Essentiality Build classifier to predict essentiality from diverse features in S. cerevisiae Apply it to S. mikatae & show performance is as good as homology Describe Network Dynamics in Yeast Reg. Net Merge expression data with net Active network markedly different in different conditions Identify transient hubs associated with particular conditions Use these to annotate genes of unknown func Gerstein.info/talks Mark Gerstein,

(JAVA) + Web service (SOAP) with Cytoscape plugin [Yu et al.

48 TopNet an automated web tool (vers. 2 : "TopNet-like Yale Network Analyzer") Normal website + Downloaded code (JAVA) + Web service (SOAP) with Cytoscape plugin [Yu et al., NAR (2004); Yip et al. Bioinfo. (2006); Similar tools include Cytoscape.org, Idekar, Sander et al] Gerstein.info/talks Mark Gerstein,

49 50 50 Gerstein.info/talks Mark Gerstein, Acknowledgements TopNet.GersteinLab.org MS MG

50 Gerstein.info/talks Mark Gerstein, Acknowledgements TopNet.GersteinLab.org MS MG

51 52 52 Gerstein.info/talks Mark Gerstein, Acknowledgements TopNet.GersteinLab.org MS MG

52 Acknowledgements TopNet.GersteinLab.org MS Gerstein.info/talks Mark Gerstein, NIH, NSF, Keck Job opportunities currently for postdocs & students A Sboner A Paccanaro M Seringhaus MG H Yu A Borneman K Yip N Luscombe S Teichmann S Douglas M Babu

53 54 54 Gerstein.info/talks Mark Gerstein,

Predicting essential genes in fungal genomes

Predicting essential genes in fungal genomes Michael Seringhaus, Alberto Paccanaro, Anthony Borneman, Michael Snyder and Mark Gerstein Genome Res. 2006 16: 1126-1135; originally published online Aug 9, 2006; Access the most recent version at doi:10.1101/gr.5144106