Sign in or Join FriendFeed
FriendFeed is the easiest way to share online. Learn more »
ISMB/ECCB
ISMB/ECCB Stockholm 2009: ISMB/ECCB
Keynote: Daphne Koller - Individual Genetic Variation: From Networks to Mechanisms
June 29, 2009 - Comment - Share
aka Understanding Gene Regulation: From Networks to Mechanisms - Allyson Lister
(as some results only arrived last night) - Oliver Hofmann
There are many mechs (chromatin modification, rna degradation) that modify gene regulation - Allyson Lister
use 'omics' data to understand and infer GRNs - Diego M. Riaño-Pachón
mRNA level of regulator an (imprecise) indicator of regulator activity - Oliver Hofmann
So you can model RNA level of a gene as a variable in a model - Allyson Lister
The expression of a target genes can often be predicted by the expression of its regulators - Diego M. Riaño-Pachón
Regulators: transcription factors, signal transduction, chromatin remodeling . . . - Diego M. Riaño-Pachón
Broad view of a regulator gene: TFs, signal transduction proteins, RNA processing factors, anything that *might* play a direct or indirect role in gene regulation - Oliver Hofmann
Second assumption: co-regulated genes have similar regulatory mechanisms, group genes into modules and predict expression profile for the entire module - Oliver Hofmann
Modules provide increased statistical power. - Gabriele Sales
A critical aspect is the structure of the regulatory program. - Allyson Lister
(Segal 2003 Nat Genetics): notion of a regression tree for regulatory programs - Oliver Hofmann
has disadvantages such as poor regulator selection lower in the tree, misses lot of regulators due to lack of statistical power - Oliver Hofmann
another disadvantage: arbitrary choice among correlated regulators - Gabriele Sales
Instead: Lasso (L1) regression approach - Oliver Hofmann
Simple linear regression doesn't work well because the minimization procedure assigns a nonzero weight to each regulator. - Gabriele Sales
This can be solved by using regularization: an extra termn penalizing nonzero weights. - Gabriele Sales
The Lasso regression pushed the effect of many regulators towards zero, likely keeping only the significant regulators - Diego M. Riaño-Pachón
Elastic net regression to avoid arbitrary regulator / feature choice - Oliver Hofmann
Used for examples in this talk - Oliver Hofmann
Cluster genes to modules, learn regulatory program for module, repeat for all modules, iterate after re-assignment of genes to modules based on how well a program predicts the expression of a gene in the module - Oliver Hofmann
initial assignment of genes into modules is often poor at the beginning, must be reiterated down in the analysis - Diego M. Riaño-Pachón
It's a special kind of Bayesian network - Allyson Lister
Test using eQTL data set (Brem, 2002 Science), two different yeast strains - Oliver Hofmann
eQTL dataset (Brem at al, 2000) from yeast. Microarray measurements of 112 individuals over 6000 genes. - Gabriele Sales
Adapt the regulatory network approach by including the genotype / markers. How do markers affect the expression level of a given module? - Oliver Hofmann
She then show us one of the modules that comes out, the telomere module (40/42 of genes are in the telomere). - Allyson Lister
Example: the telomere module. Enriched for telomere maintenance and helicase activity. - Gabriele Sales
Controlled by a region on chr XII (which includes a regulator gene) - Oliver Hofmann
23 modules out of 165 have 'chromosomal features', 16 chromatin regulators - Oliver Hofmann
Puf3 module: 147/153 genes are pulldown targets of mRNA binding protein Putf3. - Gabriele Sales
But Puf3 is not the most significant regulator of the module. - Gabriele Sales
P-bodies are places where mRNA are stored temporarily and while they are there they are transcriptionally repressed. - Allyson Lister
What regulates sequence-specific localization of mRNAs to P-bodies? - Gabriele Sales
They did a microscopy experiment to test this, which demonstrates that PUF3 is specifically localized to p-bodies. - Allyson Lister
Microscopy experiment: fluorescence shows that puf3 localizes in P-bodies. - Gabriele Sales
So what regulates the regulators of the P-bodies? - Oliver Hofmann
What regulates the p-bodies? What is one level higher up in the hierarchy? A locus on chromosome 14, but this is a large region and covers 30 genes and 300 polymorphisms. - Allyson Lister
Therefore she came up with the idea of regulatory potential. The motivation is that not all SNPs are equally likely to be causal - Allyson Lister
how to rank the polymorphisms (SNPs) by their regulatory potential? - Diego M. Riaño-Pachón
Again use the bayesian LASSO - Diego M. Riaño-Pachón
Prior distribution is a Laplacian. - Gabriele Sales
Weight can more easily deviate form 0 and the regulator is more likely to be selected. - Allyson Lister
Each regulator with its own prior determined by regulatory features of the regulator - Oliver Hofmann
it is better to give to each regulator its own prior - Diego M. Riaño-Pachón
Each regulator now has its own probability distribution about how likely it is to diverge from 0. - Allyson Lister
Metaprior Model (Hierarchical Bayes. - Allyson Lister
Each regulator has its own prior dictated by the regulatory features (inside a gene? protein coding region? strong conservation? TF binds to module gene?) - Gabriele Sales
Start by learning regulatory programs as described earlier. Second, learn regulatory weights (betas). Then compute the regulatory potential of each SNP in the genome. Then interate - Allyson Lister
Empirical hierarchical bayes - Diego M. Riaño-Pachón
Regulatory potentials do not change the selection of strong regulators, but helps to disambiguate between multiple weak regulators - Oliver Hofmann
Regulatory potentials do not change selection of strong regulators. They only help disambiguate between weak ones. - Gabriele Sales
Strong regulators teach us what to look for in the putative weak regulators - Oliver Hofmann
Some important features learned by the algorithm: conservation, cis-regulation. - Gabriele Sales
Important factors: cis-relationship, conservation, stop-codons, (combination of gene functions such as RNA modification, DNA binding, ...) - Oliver Hofmann
Statistical evaluation: uses PGV: % of genetic variation explained by the predicted regulatory program for each gene. It's a form of test data validation. - Allyson Lister
These factors are *learned* from the model, not set - Oliver Hofmann
Explained about 50% of the variation in about 50% of the genes - Allyson Lister
Back to the P-body regulation from a region on chr 14. - Gabriele Sales
Regulatory potentials are specific to organisms or even datasets. - Gabriele Sales
Understanding the process underlying differentiation with the ImmGen consortium - Oliver Hofmann
Can identify shared regulatory programs for all 60 cell types, but lumps together very different cell types - Oliver Hofmann
By using G-Regulators, you allow programs to depend on genetic variation, but you don't have G-regulators here, you have cell types - Allyson Lister
One network for each cell type overfits the data, but can bias towards shared regulation. Use the _ontogeny_ to guide conserved regulation. - Oliver Hofmann
Two extremes: lumping every cell type together you "average" effects; working on a single cell type you overfit. - Gabriele Sales
use ontogeny to guide conserved regulation. You do this by looking at differences, and penalize every place you change the regulatory power - penalize changes/divergences in the regulatory - Allyson Lister
Expression changes and underlying phenotype. What are the mechanism underlying them? - Allyson Lister
Example: transformation of FL to DLBCL occurs in 40-60% of patients, and diverse mechanisms seem to drive transformation. - Allyson Lister
DLBCL (Diffuse large B cell lymphoma) - Oliver Hofmann
Represent each module as a metagene expression profile, and use machine learning to id modules distinguishing FL-t (pre-transformation) from transformed DLBCL - Allyson Lister
Represent each module as a metagene, use ML technique to learn classifiers to distinguish FL-t (pre-transformed) from DLBCL - Oliver Hofmann
Can you use a module-based approach to understand metabolic syndrome? - Allyson Lister
pheontype network, where the nodes are modules and the edges are learned regulatory programs. - Allyson Lister
An important module in this case is the biosynthesis liver module: the genes are almost disjoint but are all in the same module, therefore you would have missed it without these modules. - Allyson Lister