15-780: Graduate Artificial IntelligenceComputational biology: Systems biology and the cell cycleMovie1.gvpThe Cell Cycle• The process in which cells divide.• Plays key role in development and cancer.Time series expression datagenesExperiments (over time)100 20 70 80gene 1Higherexpression compared to baselineLower expression compared to baselinebaseline expressionSpellman et al Mol. Biol. Cell 1998Expression = level of gene (protein) in this experimentTime line• 1997, 1998 – budding yeast cell cycle expression• 2000 – plants • 1999, 2000 - human• 2001 – mouseTime line• 1997, 1998 – budding yeast cell cycle expression• 2000 – plants • 1999, 2000 - human• 2001 – mouse• 2002 – Human data is noise !Reproducibility of peak between two repeatsTime line• 1997, 1998 – budding yeast cell cycle expression• 2000 – plants • 1999, 2000 - human• 2001 – mouse• 2002 – Human data is noise !• 2002 – Cancer cell cycle expression (approximation)Time line• 1997, 1998 – budding yeast cell cycle expression• 2000 – plants • 1999, 2000 - human• 2001 – mouse• 2002 – human data is noise !• 2002 – cancer cell cycle expression (approximation)• 2004, 2005 – deconvolution and Checksum• 2006 (upcoming) – human cell cycle dataTime line• 1997, 1998 – budding yeast cell cycle expression• 2000 – plants • 1999, 2000 - human• 2001 – mouse• 2002 – human data is noise !• 2002 – cancer cell cycle expression (approximation)• 2004, 2005 – deconvolution and Checksum• 2006 (upcoming) – human cell cycle data • 2004 – fission yeast cell cycle data“Our comparisons with budding yeast data revealed a surprisingly small core set of genes that are periodically expressed in both yeasts.”• Cells are synchronized to the same phase• Microarray experiments at multiple time points after release from synchronization• Scores derived from multiple expression time series• Rank genes based on their scores, and use a cutoff score to identify cycling genesFrom expression values to scoreSpellman et al. (1998)Problems• Different scoring methods result in different lists• Microarray data are noisy• Hard to separate scores for cycling and non-cycling genes– Score distribution of cell cycle genes (derived from GO) versus the rest• solid curve: cycling genes• dotted curve: the restSignificance of homolog overlapGraphical models• Efficient way to represent and reason about joint distributions• Graphs in which nodes represent random variables and edges correspond to dependency assumptions • Two major types: Directed and undirected)](|[iiixPaxp∏• Bayesian networks• Hidden Markov models∏jijijixx,,),(ψ• Markov random fieldsGraphical models (cont)• Parameters are used to specify the conditional probability distribution (directed graphs) or the potential functions (undirected graphs)• Computational questions:- Determining the structure of the model (sometimes)- Estimating the parameters of the model- InferenceProbabilistic graphical model for combining expression and sequence homology… …… …… …: Ci: Cycling Status Nodes (unobserved): Encodes Dependency Relations: Si: Score Nodes (observed)… …… …Numeric summary of expression time series• Node Potential:• Edge Potential: • Joint probability distributionLikelihood of the model… …… …… …… …… …need to be learned from dataweight (from homology)controls contribution from each sourceIn what cases can we improve? Arabidopsisbudding yeasthumanfission yeast051015202530354045all cycling inbuddingcycling, budingonlyconserved insequence conserved insequence andexpressionPercentageSeries1Essential yeast genes10-810: Graduate Computational Genomics • Spring 2007• Bar-Joseph, Benos, Xing• TR 10:30-11:50, Scaife Hall, 208• Intro to computational biology emphasizing machine learning, sequence analysis and systems
View Full Document
Unlocking...