 |
|
| FriendFeed is the easiest way to share online. Learn more » |
arne › Likes
Surehy Breter, Samuli Eldfors, Ruchira S. Datta and 4 other people liked this
George Church has developed an amazing amount of technology.
- Barb Bryant
I am always wondering that if he gets any sleep at all.
- Dawei lin
Which is the introducer?
- Dawei lin
michal linial, if I'm not wrong (which I was, need new glasses)
- arne
The My First DNA sequencer reference: http://www.wired.com/wired...
- Shannon McWeeney
First challenge on computational interpretation and integration: personal genomes =stem cell epigenome + mC environments + traits.
- Dawei lin
Olga Troyanskaya
- Barb Bryant
Cost of drugs goes up linearly; cost of sequencing is dropping exponentially
- Barb Bryant
40,000 fold price drop for 4 years
- Dawei lin
CGI price for genome is $1500/year?
- Dawei lin
In 2005 we abandoned a monopolistic capillary electrophoresis; instead we have a couple and now 21 different technologies for sequencing. Resulted in a jump in rate of change of sequencing capacity
- Barb Bryant
He thinks that many of the sequencing companies will find a niche :)
- arne
Cost of personal genome: 2007: $57M; 2009 $1500, for 40-fold coverage.
- Barb Bryant
Close to the $1000 genome
- arne
(+ $100,000 interpretation cost?) (he doesn't really think that)
- Barb Bryant
Drmanac et al Science Jan 2010
- Dawei lin
Sidetrack: One friend said when he started his PhD it took 6 month to sequence a bacteria and 6-60 month to analyse it. Not it takes 6 minuted to sequence it and still 6-60 month to analyze it.
- arne
limitation is several hundreds nm in scale on chip (positive charge molecules on hydrophobic background
- Dawei lin
7% human genome is missing so far because of technical challenges
- Dawei lin
trio genomics information (father, mother, child) is increasing important in genomics research
- Dawei lin
From open acess Sequences to Bio-Fab
- arne
One of the 21 sequencing technologies is open-access. Reads and writes DNA with light.
- Barb Bryant
2nd-gen synthesis ($500 per 15 Mbp)
- arne
Second-generation synthesis - four different kinds of technologies.
- Barb Bryant
Next Gen synthesis: off chips $500 15Mbp
- Dawei lin
Tian et al 2004 Nature
- arne
The work started around 2003
- Dawei lin
personalgenomics.org
- arne
person genome 3M allele -> immunology + microbome -> trait
- Dawei lin
Issues of personal identification from genomic data. Informed consent as one solution.
- Barb Bryant
Have 16,000 volunteers for Personal Genome Project so far; 100,000 target.
- Barb Bryant
Claims that ~1800 genes are highly predictive and medically actionable.
- Barb Bryant
They are rare but collective common at 10% level
- Dawei lin
Example of the Madsen family with two diseases. Found causative allelles - 4 total (2 from each parent).
- Barb Bryant
evidence.personalgenomics.org
- Dawei lin
Each time we find a scary allele in a person, it could be a sequencing error; it could be a problem with the literature.
- Barb Bryant
found a dozen cases in the literature got allele sequence wrong
- Dawei lin
The oldest volunteer for PGP is 96.7 years old
- Dawei lin
Q: Are these genomes available ?
- arne
Circulating tumor, pathogen, fetal, and immune cells.
- Barb Bryant
Microbe vs Immunome
- arne
If you want to look for a microorganism in a body, you can either look directly for the microbe, or look for the body's reaction.
- Barb Bryant
immune test is to focus on response to exposure.
- Dawei lin
Sequencing after vaccination - response is maximum after 7 days
- arne
Generating human tissue from pluripotent stem cells
- Barb Bryant
The Economist 20-May-2010 cover
- Dawei lin
Genome engineering
- Barb Bryant
E.g., change the genetic code -- for resistance to pathogens, new amino acids, and something else.
- Barb Bryant
You have to do this safely.
- Barb Bryant
For $400M, Dupont made 27 changes to the 4.6 Mbp E. coli, to make a chemical.
- Barb Bryant
Another application: bio-petroleum from microbes.
- Barb Bryant
Identify enzymes that synthesize alkane. Many cyanobacteria made trace amounts; others made none. Did genome sequence "subtraction" to find which genes were in the former. Isolated & tested these genes. Overproduced them; it worked. Green chemistry.
- Barb Bryant
Multiplex Automated Genome Engineering (MAGE)...
- Barb Bryant
Church's own genome is available: http://www.personalgenomes.org/public...
- Christiaan Klijn
So: subtract my genome from Church's, then overproduce those genes --> TOTAL BRILLIANCE!
- Barb Bryant
Example of freeing up a codon by changing those codons to a different one./
- Barb Bryant
Is this not just the analysis. Not the sequence ? (or did I miss a link)
- arne
See the 'Datasets' header -> you can get 500k Affy data as well as exome
- Christiaan Klijn
Metabolic engineering example. Historically, you'd get obsessed with one step in the pathway and overproduce one enzyme. But then you'd get product inhibition, or the product might be toxic.
- Barb Bryant
Would be nice with a map to the reference genome as well, but guess that can be done
- arne
DNA Nanostructures: (DNA origami). Proposes a combination of DNA and proteins.
- arne
DNA nanostructures help solve structures of membrane proteins.
- Barb Bryant
First practical application: Made a long rod that was stiffer than other DNA. Used in NMR for membrane proteins (Cooooll idea but, it has been tried with proteins before)
- arne
caDNAno is a software tool that is free available
- Dawei lin
Time for questions.
- arne
Roland Krause, Venkata P. Satagopam, Mickey Kosloff and arne liked this
Shows picture of stages of cancer progression (ref Vogelstein, colon); poses the question of how metastasis occurs -- does this involve genetic or epigenetic changes?
- Barb Bryant
Tan Ince cultured two kinds of normal human mammary epithelial cells. He transformed them with oncogenes, resulting in different types of tumors.
- Barb Bryant
Concludes that the nature of the normal cell of origin is a strong determinant of the phenotype of the primary tumor, and whether it metastasizes. The playing field is tilted in the beginning.
- Barb Bryant
Posits tumor-generating cells.
- Barb Bryant
Self-renewing stem cells produce either more stem cells or transit amplifying cells which in turn lead to post-mitotic differentiated cells. Only the self-renewing stem cell could seed a new tumor.
- Barb Bryant
invasion-metastasis cascade
- Barb Bryant
How do cancer cells acquire all of these capabilities (invasion, intravasastion, transport, metastasis...) Are there addiitonal mutations required? Is it epigenetic?
- Barb Bryant
epithelial-mesenchymal transition -- cells on the perimeter of the tumor are mesenchymal. This may be due to signals from the surrounding stroma.
- Barb Bryant
There are probably 1000 proteins that shift in EMT. Vaious transcription factors (TFs) induce EMTs.
- Barb Bryant
EMT program highly complex and occurs normally during development.
- Mickey Kosloff
from iPod
It seems likely that most of the invasion-metastasis program can happen without need for additional mutations; rather use signaling from microenvironment.
- Barb Bryant
P. Gupta transformed human primary melanocytes (pigmentation in the skin) with a cocktail of oncogenes. Found that in contrast to transformed epithelial cells, there was much higher likelihood of metastasis. Again, cell of origin is important in future behavior.
- Barb Bryant
One TF, Slug, was found to enable melanoma metastasis. (Even though the primary tumors grew a little faster.)
- Barb Bryant
Another TF, FOXC2, when expressed in epithelial cells induces migration and invasion. A subset of breast cancers have high levels of nuclear FOXC2, and these are more aggressive breast cancers.
- Barb Bryant
Speculates that different networks of EMT-inducing factors might program metastasis in different cell types./
- Barb Bryant
Stem cells identified by high CD44 and low CD24. (CD's are markers on cell surface which can be assayed fairly easily.)
- Barb Bryant
There are various ways to make cells acquire stem cell characteristics.
- Barb Bryant
Mentions Kornelia Polyak. There are stem-like cells in primary human breast samples. The stem cell program in normal human mammary gland is coopted by cancer cells.
- Barb Bryant
More proof that EMT creates stem cells.
- Barb Bryant
Most current chemotherapies preferentially kill non-cancer-stem-cells. The remaining stem cells can repopulate the tumor and are often more resistant to therapies.
- Barb Bryant
Gupta & Onder tested CSCs and non_CSCs with a bunch of drugs. There are some CSC-targeted agents (Salinomycin, Abamectin). Of 16,000 compounds only about a dozen preferentially killed CSCs as opposed to non_CSCs. Many were the other way round.
- Barb Bryant
This probably won't be the "answer". Christine Chaffer noticed that there were some floating cells in 2D cultured human mammary epithelial cells. She grew these up; these look more like CSCs.
- Barb Bryant
Interestingly, she found that non-CSCs could generate CSCs.
- Barb Bryant
Hm, isn't this kind of pouring cold water on the excitement about CSCs as drug targets? Or maybe you have to target both CSCs and non-CSCs simultaneously.
- Barb Bryant
yup
- Barb Bryant
Q: cancer biologists like to study druggable genome. But transcription factors seem most important. A: expression of TFs is controlled by cytoplasmic factors. Might want to go after those. Drugging the TF itself might be hard, but the signaling pathways might be more druggable.
- Barb Bryant
Q: has it been shown that change in the two forms of cadherins match the change in CD expression, and are these correlated with morphology? A: I showed that: CD44 high cells shut down E-cadherin; they expression vimentin, and other mesenchymal markers. I don't know whether CD44 is useful for non-mammary epithelial tissues.
- Barb Bryant
Q: So do normal non-SCs generate SCs? A: Yes. Same differences as in cancer.
- Barb Bryant
Spontaneous de-differentiation into SCs. Interesting phenomenon.
- Steve Chervitz Trutane
arne and Roland Krause liked this
Omic data don't mean what you think
- Roland Krause
Generally little overlap between different experimental screens.
- Roland Krause
Surehy Breter, Guilhem Chalancon, Ruchira S. Datta and 2 other people liked this
p110-p85 complex
- arne
Similar to Lego assembly
- arne
combine different types of data/constraints to solve the docking problem
- Mickey Kosloff
used patchdock for rigid docking
- Mickey Kosloff
Patchdock: http://bioinfo3d.cs.tau.ac.il/PatchDo...
- arne
SAXS: Using a FFT can provide distance between atoms (on average)
- arne
3D-EM: Density map. Superposition to model. fit score
- arne
2D-EM: Similarity
- arne
Fast mapping by NMR spectroscopy
- arne
FIREDOCK (a method for fast flexible refinement) http://bioinfo3d.cs.tau.ac.il/FireDoc...
- arne
Example: Antigen - antibody target..
- arne
Hmm.... my posts never appeared !!
- arne
had strange delay problems during keynote
- Mickey Kosloff
JaraBreter, Roland Krause, Mickey Kosloff and 3 other people liked this
Altshuler is an expert on diabetes type II.
- Dawei lin
It is said that he is also a good dancer.
- Dawei lin
Tap, ballroom, or tango?
- Ted Laderas
Slide dancing
- Dawei lin
motivation is to understand genetic basis of human diseases
- Dawei lin
Genetic basis of human diseases - important disease mechanisms and bio pathways remain unidentified
- Venkata P. Satagopam
gap in knowledge of human disease biology contribute to high failure rates in drug development
- Dawei lin
Why understanding genetic mechanisms ? (1) Important mechanism remain unidentified (ii) Gaps in knowledge causes failure rate in drug development
- arne
It will be a long way to know if the two motivating hypotheses are true
- Dawei lin
one of the most research on T2D. It scaned 100k people for 10 yrs
- Dawei lin
10 years later 50% progressed to have the disease
- Dawei lin
10years of diabetic research - the out come is - 50% of people with good lifestyle improved
- Venkata P. Satagopam
lifestyle has a bigger impact than Metformin
- Dawei lin
Diabetes study with 10-year follow-up of diabetes incidence and weight loss, "T2D". Randomized into treatments: lifestyle, metformin, placebo. Best drug makes relatively little difference in incidence; lifestyle intervention is better than drug but still doesn't help a whole lot.
- Barb Bryant
best prevention was extensive lifestyle changes (50% -> 40% incidence)
- Mickey Kosloff
Diabetes is not only a matter of life style
- arne
success rate in current pharma industry is <5% of molecules entering the clinical trails
- Venkata P. Satagopam
This is bad !!
- arne
mentions well known number of >95% failure rate of new compounds
- Mickey Kosloff
because there are still 40% people got the disease after the lifestyle change, it seems that people do not know the course of the disease
- Dawei lin
a genetic approach - http://www2.massgeneral.org/chrg...
- Venkata P. Satagopam
Genetic mapping started in 1913
- Dawei lin
genetic map came in 1913
- Venkata P. Satagopam
Morgan and Sturtevant 1913
- arne
emphasizes he advocates a genetecist's approach (rather than a genomic approach)
- Mickey Kosloff
And tells you to skip undergraduate work if you have something better to do
- arne
key attributes of genetic mapping - unbiased by prior assumptions about pathways
- Venkata P. Satagopam
saturation mutagenesis reveals pathways
- Venkata P. Satagopam
key attributes of genetic mapping: (1) unbiased by prior assumptions about pathways (2) saturation mutagenesis reveal pathways
- Dawei lin
many mutants -> reveals coherence of pathways
- Ted Laderas
These days we have other methods that are unbiased like expression profiling, but genetic mapping has some unique characteristics relative to these (he’ll explain in a minute).
- Barb Bryant
Drosophola's mutations looked initially random, years they almost all related to pathways.
- Dawei lin
bottleneck is functional determination - biochemical approaches
- Ted Laderas
A lot of current knowledge can track back to genetic mapping
- Dawei lin
Botstein and Fink Science 1988 ....
- Venkata P. Satagopam
A slide based on Galzier et al, Science 2002
- Dawei lin
genetic mapping of human single gene disorders ...over 15 years Botstein paper in 1980, first genetic map in 1985 ....
- Venkata P. Satagopam
It took 10 year to find maker for Huntington disease
- Dawei lin
Once you find a linked region from genetic mapping, it still takes a long time to find the specific gene responsible.
- Barb Bryant
in the 1990's the idea was that common diseases were caused by rare mutations with large effects
- arne
"Chromosome shlepping" - Eic Lander's term for the identification of a very gene in some genomic region.
- Roland Krause
It is robust to find mendelian disease but to not common diseases
- Dawei lin
another approach: population genetics - QTL approach
- Ted Laderas
phenotypic variation is often continuous and may involve variation in many genes
- Dawei lin
Galton invented regression analysis to analyze the measuring of phenotypic data (heights of parents and offspring).
- Roland Krause
The biometric unit --- almost nothing was Mendelian
- arne
Most traits are continuously variable
- Ted Laderas
Francis Galton was a cousin of Darwin. Darwin didn’t explain the source of variation. Galton focused on this; he measured the heights of parents and their offspring, and found a relationship. He invented regression analysis to draw the line. The slope of the line is related to the inheritability of the disease.
- Barb Bryant
It was studied by the cousin of Darwin, Francis Galton (1885)
- Dawei lin
phenotypic variation is often continuous ... some history ... Francis Galton (1885), Ronald Fisher (1918), Hermann Muller (1920)
- Venkata P. Satagopam
This gave rise to the biometric movement – measure every living thing. Traits were related to genetic relatedness; and it wasn’t Mendelian. This led to the biometric-Mendelian debate.
- Barb Bryant
Ronald Fisher, was actually a geneticist, who also invented p-value and Fisher exact test
- Dawei lin
Ronald Fisher (the one with the exact test) was also a geneticist.
- Roland Krause
Solved by assuming that phenotype often is an effect of several Mendelian genes.
- arne
Fisher: individual genes are mendelian, effects of genes additive
- Ted Laderas
Hermann Muller 1920 (Nobel Prize for X-ray induced mutations). PhD thesis not Mendelian trait, but truncate wing. Wasn’t Mendelian. Did genetic mapping.
- Barb Bryant
Hermann Muller decided to use broken wing of fruit fly to study non-Mendelian diseases
- Dawei lin
Muller 1920 paper: 4 chromosomes in fly – 3 contain genes that influence the trait truncate wing. Muller wrote about implications for human traits, like psychological traits. Said that traits were going to be too complicated. Said you could figure out by looking at population, but not looking at Mendelian inheritance in families.
- Barb Bryant
Muller 1920 suggested that it needed to do study on a population.
- Dawei lin
Muller: Truncate wing - 3 genes influence effect of phenotype
- Ted Laderas
Mullers thesis included the notion of surveying complex phenotypes in the population rather than families.
- Roland Krause
Muller: traits are too complex to observe in families, but can observe in population
- Ted Laderas
characterization and catalogue human seq variation is a decade of work .. i.e international HapMap project
- Venkata P. Satagopam
Another decade-long failure: the candidate gene approach. Instead, we need a genome-wide, unbiased approach.
- Barb Bryant
Testing candidate genes was not successful. Only 10-20 successes.
- Dawei lin
779 GWA published for 148 traits
- Mickey Kosloff
out come - 779 published GWA for 148 trails
- Venkata P. Satagopam
For common diseases, GWA was needed
- Ted Laderas
but "correlation does not imply causality"
- Mickey Kosloff
There have been 779 genome-wide association studies (or regions/genes found?) for 148 traits, with p < 5x10^-8
- Barb Bryant
"correlation does not imply causality" ....
- Venkata P. Satagopam
But correlation does not imply causality.
- Barb Bryant
The reasons of "Correlation does not imply causality": irreproducibility, lack of randomization, confounding, arrow of time.
- Dawei lin
If you can't randomize the experiment you can never prove causality as opposed to just being correlated to the underlying cause.
- Barb Bryant
FF lag results in all these duplicate posts
- Mickey Kosloff
a lot of efforts are on finding correlation between rare variation and diseases
- Dawei lin
rare variation is defined as has <5% in population
- Dawei lin
95% of variations is already present in the database
- arne
Identified 50 regions that are associated with T2D
- arne
with in next few years ... the role of rare and less common variants will be characterized in a variety of diseases
- Venkata P. Satagopam
next topic - can we obtains new insights into the basis of disease?
- Venkata P. Satagopam
one example - sickle cell anemia
- Venkata P. Satagopam
Sankaran et al Science 2008
- Venkata P. Satagopam
Lettre et al PNAS 2008
- Venkata P. Satagopam
Uda et al PNAS 2008
- Venkata P. Satagopam
Crohn's disease: 15 years, no idea what was happening. Now many genes and 3 pathways are identified to be relevant.
- Dawei lin
96 loci explain ~25% of cholesterol levels
- Mickey Kosloff
Lipid GWAS found 60 loci that are previous unknown. Some of the positives are drug targets already.
- Dawei lin
Global lipids consortium, forthcoming Nature paper (Nature paper is mentioned about 20 times !!!)
- arne
is there a way to automate validation/function determination?
- Ted Laderas
prediction -- will prediction prove useful --this is depending on the clinical testing and the genetic test
- Venkata P. Satagopam
prediction will be useful when there's a proven intervention
- Mickey Kosloff
BRCA1/2 risk for cancer as an example
- Mickey Kosloff
seq tech will increase the reach of genetic methods
- Venkata P. Satagopam
mendelian fallacy - sub-populations are easily divisible in terms of risk
- Ted Laderas
Prediction will only be useful if there is an intervention that you would not use without the prediction. Otherwise, you should use the intervention anyway.
- Roland Krause
Huntington will not be a representative example - for most diseases/people identified risk will be <<100% even with full genetic information
- Mickey Kosloff
Cautionary tale - PSA prediction results in over-treatment, hasn't been shown that people live longer because of test
- Mickey Kosloff
Very cautious about PSA - no improvements on the mortality but many operations performed.
- Roland Krause
genetics offers a path to discover the underlying biology of human diseases ; the great value will drive from pathophysiology and treatment
- Venkata P. Satagopam
arne, Mickey Kosloff, Saravanamuttu Gnaneshan and 3 other people liked this
An interview with Chris Sander ... http://www.mskcc.org/mskcc...
- Venkata P. Satagopam
Kabsch and Sander paper - over 6000 citations - http://www.ncbi.nlm.nih.gov/pubmed...
- Shannon McWeeney
Note the subliminal message in the announcement slide
- Iddo Friedberg
from Android
Prediction by transparency - no computation necessary story
- Shannon McWeeney
Awards should be shared: People working with Chris includes: Burkhard Rost, Alfonso Valencia, Liisa Holm and many more
- arne
Announcement of unpublished and new work. A good trend at this ISMB.
- Roland Krause
Cancer genome atlas: TCGA
- arne
Mapping of molecular alterations (cpy number variation) to 200 glioblastoma samples. http://www.ncbi.nlm.nih.gov/pmc...
- Roland Krause
Difference between patients is huge
- arne
extract network, find relevant modules.
- Roland Krause
illustration of netbox algorithm
- Shannon McWeeney
When grouping mutations into pathways up to 85% of GBM have a muation in the most important pathways, while individual genes are down to a few %
- arne
Each oncogene may have relatively low frequency across patients; but when you group genes across pathways, a pathway may explain a large fraction of patients with a given type of cancer.
- Barb Bryant
"Network pharmacology"
- Barb Bryant
can see a change in pathway activation between primary tumor and mets
- Mickey Kosloff
Dominant alterations changes between cancer types and states.
- Roland Krause
GBM: copy number is rare (and noisier) Ovarian: more regular and higher
- arne
profiles of copy numbre variations differ between types of cancers
- Mickey Kosloff
Metastatic tumor samples have more copy number changes than primary tumors. Not surprising. But maybe primary samples with more copy number changes than others are more likely to metastasize? Generally, better outcome with fewer somatic copy number changes.
- Barb Bryant
BRCA1 and BRCA2 mutations convey germline inherited cancer risk
- Barb Bryant
These genes act in the homologous repair pathway. Half of all patients have mutations in some homologous repair pathway gene.
- Barb Bryant
and more generally, homologous repair genes are altered in > 50% of ovarian cancer
- Mickey Kosloff
Tumor suppressor genes can be inactivated in various ways: germline mutation, somatic mutation, epigenetic silencing, etc.
- Barb Bryant
There are drugs under development that might work particularly well in patients with defects in this particular pathway.
- Barb Bryant
Cancer genomics portal: www.cbio.mskcc.org/cancergenomics
- Barb Bryant
mutationassessor.org
- Barb Bryant
Topic shift: now, perturbation cell biology. "and belief propagation". (eh?)
- Barb Bryant
Perturbation Cell Biology
- arne
In recent past, says Chris, you make a few perturbations: overexpress or knock down a gene; inhibit with a compound, etc.
- Barb Bryant
use network inference algorithms
- Mickey Kosloff
goal = predictive models for therapy
- Mickey Kosloff
with only 200 datapoints -> derive validated (known) pathways
- Mickey Kosloff
Prediction of networks does not scale to larger networks
- arne
Large data generation with the number of pertubation > than proteins.
- Roland Krause
Still prohibitively large number of networks even for small number of nodes.
- Roland Krause
Use statistical physics methods to tackle combinatorial explosion of possible networks.
- Barb Bryant
Inference using belief propagation known from statistical physics.
- Roland Krause
Ah, here is where "belief" comes in. Network inference using belief propagation. Reference Riccardo Zecchina et al. http://users.ictp.it/~zecchi...
- Barb Bryant
Instead of going through all the models that are possible, you derive statistical properties across a set of good models for each of the Wij weights in the model.
- Barb Bryant
This is sort of like partition functions in statistical physics
- Barb Bryant
evolving work on Wij (transition from Nelander et al 2008- http://www.nature.com/msb...)
- Shannon McWeeney
Cavity approach - optimize locally on global background iteratively cover all local cavities
- Shannon McWeeney
Mm, this is rather opaque to me.
- Barb Bryant
"Let me give you some intuition about how this all works." Yes, I'd like that.
- Barb Bryant
Nice results on toy experiment - constraints from 10 experiments with 5 interactions (the nodes W in factor graph).
- Shannon McWeeney
after step 1 - generation of probability distributions then step 2- decimation
- Shannon McWeeney
So you have a probability distribution for each Wij, which represents the interaction between element i and element j. I'm not really getting how you "update" these probability distributions in the iterative steps. I do understand that at the end you take the most "certain" (narrowest) distribution and fix its value (some Wij) at the most probable value, then update all the other Wij's given this fixation. And so on. To get your final model in a sort of greedy fashion.
- Barb Bryant
And by the way, the underlying model is a simple differential equation sort of thing: change of one variable xi is a sigmoidal function of weighted (Wij) sum of all variables xj, less a decay term.
- Barb Bryant
thanks for the summary bb
- Michael Jones
Mike!
- Barb Bryant
Mentions bunches of other stuff in passing. Like bioPAX: paper in press.
- Barb Bryant
bioPAX is community project on pathways, ontology, and exchange format.
- Barb Bryant
"no science without people; science for the people; ask good questions"
- Shannon McWeeney
Question: Interacting network tend to be modular, with strongly-interacting subnetworks that interact weakly with each other. ...
- Barb Bryant
Chris: Is the modular approach really useful in confronting the data? [Is that what he said?]
- Barb Bryant
Question: can you get at causal relationships?
- Barb Bryant
Chris: yes - if the network model allows you to predict correctly the result of a particular perturbation applied to a particular node, then you can simulate using that model.
- Barb Bryant
Question: with a big network, how many experiments will you need to model?
- Barb Bryant
Chris: Good question. Could use an entropy measure. Help us figure this out. Help us design the experiments. It's important because of the costs of experiment. This is going to be broadly applicable in cell biology.
- Barb Bryant
bb - he said one should see if approach is useful by confronting with real data
- Shannon McWeeney
from BuddyFeed
Ah, thx
- Barb Bryant
Chris gets at the difference between a model that tells a story and a model that is truly predictive.
- Barb Bryant
Question: yes, but, what are the semantics of the graph? What kinds of interaction? Answer: The semantics are in the mathematics of your model.
- Barb Bryant
Question: mean field approach is interesting. Compared to Monte Carlo approach, you are assuming some decoupling. Loss of posterior coupling between weights - is that an issue?
- Barb Bryant
Chris: If you look at a coupled system overall, the extent to which the algorithms work depends on correlations within the system. Long-range (in terms of network distance) correlations are problematic. There are some clever approaches to handle some of this. Mentions non-ergotic space; deal with parts of space separately or iteratively.
- Barb Bryant
Dina Schneidman, Mickey Kosloff, Tomasz Puton and 3 other people liked this
Protein Folding Requires Crowd Control in a Simulated Cell Benjamin R. Jefferys⁎, Lawrence A. Kelley and Michael J. E. Sternberg J. Mol. Biol. (2010) 397, 1329–1338
- arne
Mickey Kosloff, arne and Nir London liked this
Use protein design to understand protein space.
- Mickey Kosloff
Model designs with rigid BB (only side-chains) using rotamer library
- Mickey Kosloff
scoring usually done with pseudo-physical atomic energy + pairwise decomposition
- Mickey Kosloff
use probabilistic approach to bypass the NP hard problem
- Mickey Kosloff
One protein, two structures.
- arne
Original paper: http://www.ncbi.nlm.nih.gov/pubmed...
- Mickey Kosloff
2 examples - g-protein beta subunit and calmodulin
- Mickey Kosloff
Dina Schneidman, Ruchira S. Datta, KEJIE LI and 4 other people liked this
Published paper here: http://www.ncbi.nlm.nih.gov/pubmed...
- Mickey Kosloff
Protein structure search - a computationally hard problem
- Mickey Kosloff
Looking for similar structures that are not easily found using SCOP and CATH.
- Mickey Kosloff
1st line tool - structural alignments. However, these are expensive computationally, not feasable for full PDB. Alternative - filter methods that can look at full-size PDB.
- Mickey Kosloff
disclaimer of un-objectivity - I'm a co-author on related paper with Rachel (Kosloff & Kolodny, Proteins 2008) that she just mentioned as motivation to look at the whole PDB rather than a non-redundant subset.
- Mickey Kosloff
FragBag = new filter method, based on library of fragments
- Mickey Kosloff
order of fragments in structure is not considered (similar to how Google indexes web pages). You lose information but gain speed dramatically. Worked for Google, so might work for structures.
- Mickey Kosloff
Uses SAS score as similarity measure = RMSD*100/length
- Mickey Kosloff
Uses SAS to find best of six different structure alignment methods as gold standard.
- Mickey Kosloff
Near neighbors: SAS <= 5A
- Mickey Kosloff
checks if FragBag finds nearest neighbors found by gold standard. analysis with ROC curves.
- Mickey Kosloff
rank of methods by area under ROC curve: sequence alignment is worse (as expected) structural alignments range from > 0.7 to 0.9. FragBag does as well as CE etc., even though it's a lot less expensive in computational resources.
- Mickey Kosloff
FragBag also finds pairs of structures with same CATH classification
- Mickey Kosloff
Additional features: enables to query PDB with combination of non-continuous sub-structures
- Mickey Kosloff
Also enables to visualize protein structure space (shows rotating 3D projection of 30,000 structures) - you get the (known) separation of structures classes (alpha+beta, alpha, beta, alpha/beta)
- Mickey Kosloff
superimposes SCOP folds on this picture, visualizes co-localization of these folds.
- Mickey Kosloff
answer to Q: Can search entire PDB on laptop very quickly.
- Mickey Kosloff
arne liked this
LAST, like BLAST but faster. Handles repetetive regions and A+T bias much better than blast Blast etc used fixed seed length (Last uses a adaptive length)
- arne
Second tool: RECOUNT. 1% error per position.... Genomes are repetetive....
- arne
Third tooL SLIDESORT (see poster)
- arne
Change of speaker.....
- arne
Barb Bryant, arne, Mickey Kosloff and Roland Krause liked this
This will probably be a very interesting talk. Just can't wait.
- Tomasz Puton
Not just interesting, but most likely great. Svante is a fantastic speaker
- arne
If you’re interested in human history, the genome is a great source of information. To reconstruct history, we compare sequences of people (and other species) living today. We use models of how DNA changes over time to understand the differences that exist today. This is an indirect way to study history, because we are reconstructing from the present what we think has happened in the past.
- Barb Bryant
specimens are highly contaminated, ....
- Venkata P. Satagopam
mtDNA - advantage of many copies per cell
- Mickey Kosloff
original work from 1984 on egyptian mummy - http://www.sciencedirect.com/science...
- Shannon McWeeney
Replacement (out of africa theory) vs assimilation (i.e. geneflow from modern humans)
- arne
mtDNA is extracted from a specimen from neanderthal
- Venkata P. Satagopam
Started with the original neanderthal specimen
- arne
The variation in human population origins before the split (as measured by mtDNA) of modern and neanderthals
- arne
extract dna from skull, skip PCR and directly sequence
- Mickey Kosloff
only 3.5% actually from neanderthal genome
- Shannon McWeeney
Average length 50 nucleotides
- arne
Vindija Cave, Croatia .... 3 bones
- Venkata P. Satagopam
only about 3.5% of dna came from human
- Mickey Kosloff
3 billion fragments - again most from bacteria
- Shannon McWeeney
most dna is bacterial contaminants
- Mickey Kosloff
avg genome cover is 1.5X
- Venkata P. Satagopam
most DNA extracted is female look at Y chrom % as contaminant
- Ted Laderas
Three females samples (and therefore Y chromosome contamination can be used to calculate noise). Total risk is below 1% risk of contaimination
- arne
at any particular position - 1% chance contamination (broken down by source - 3 measures)
- Shannon McWeeney
consistant nucleotide chemical changes at 5' and 3' ends
- Mickey Kosloff
try to correct by alignments to human and chimpanzee genomes
- Mickey Kosloff
Details on bioinformatics and alignment issues (led by Ed Green) can be found in Science paper - http://www.sciencemag.org/cgi...
- Shannon McWeeney
55% chance of seeing a position covered by at least 1 read
- Ted Laderas
Divergence to human reference genome 12% highest among human is in San 10%
- arne
typical european (French) 8% divergence to human reference compared with 12% in neanderthal
- Shannon McWeeney
78 amino acid substitutions ... a catalog of novel fixed features in the human genome
- Venkata P. Satagopam
But this number will change
- arne
novel fixed features in human genome - 78 aa substitutions (in paper) - now down to 50
- Shannon McWeeney
Three out of six proteins with 2 changes are skin expressed
- arne
next focused on SNPs
- Mickey Kosloff
detection of selective sweeps - look for snps in human, chimps, neanderthals - r egions where neanderthal looks all ancestral.
- Shannon McWeeney
S vs cM plot - visual inspection for widest spread
- Shannon McWeeney
Most extreme case in THADA, http://www.genecards.org/cgi-bin... Transport and diabeted related
- arne
Thada is risk allele for type 2 diabetes - implications for metabolism
- Shannon McWeeney
detection of insertion in intron in Thada (not fixed in humans as initially thought in paper)
- Shannon McWeeney
3-4% in europe has the neanderthal version (and are protected against Diabetes Type II)
- arne
interesting follow-up research here - positive selection yet cost with risk allele
- Shannon McWeeney
RUNX2: Mutations cause CCD (Cleidocranial dysplasia)
- arne
annotation of others associated with autism and other diseases including CCD
- Shannon McWeeney
CCD of interest due to skull morphology phenotype
- Shannon McWeeney
Now comes the most surprising result.
- arne
focusing on - Interbreeding with modern humans?
- Venkata P. Satagopam
Work by Rasmus Nielsen http://ib.berkeley.edu/researc...
- arne
Is Craig Venter a "fully modern human" ?
- arne
analysis of self-identified neanderthals who write to Svante - predominantly men.
- Shannon McWeeney
Comparisons to genomes of humans from different continents suggests interbreeding occured in middle east, before geographic expansion
- Mickey Kosloff
:)
- arne
45% men who are neandertals, 1% women are neandertals....
- Venkata P. Satagopam
future 10-20x coverage of genome
- Mickey Kosloff
Future: (i) Better coverage (10-20x coverage) (ii) Functional analyses of candidate genes Exemplified by FoxP2 http://en.wikipedia.org/wiki...
- arne
next topic - functional analysis of genes - foxp2
- Venkata P. Satagopam
FoxP2 is the same in human and neanderthal.
- arne
hope to identify backmutations in humans -cheaper to find these people because of low cost of sequencing
- Ted Laderas
easier to check phenotypes in mice
- Mickey Kosloff
Human FoxP2 in mouse: The mouse can not speak ! Large scale phenotype study (323 phenotypic traits). -> More cautious in a novel area (stays close to the wall). No difference after 3 minutes. Second phenotype: Altered vocalization !!!
- arne
323 phenotypic traits ... studied ..
- Venkata P. Satagopam
movement more cautious in humanized mice
- Venkata P. Satagopam
next one is altered vocalization
- Venkata P. Satagopam
Enard et al Cell 2009
- Venkata P. Satagopam
mice with human foxp2 grew longer neurons
- Mickey Kosloff
Other hominid forms........
- arne
Denisova individual 1 Myears (400 diffs in mtDNA)
- arne
very good keynote
- Mickey Kosloff
Barb Bryant, Mickey Kosloff, Deepak Singh and 4 other people liked this
ISMB2010 just kicked off
- Venkata P. Satagopam
Prof Søren Brunak introducing Steven Brenner, ISCB overton prize winner
- Venkata P. Satagopam
Brenner contributed to many fields in bioinformations, starting in structureal biology ober RNA to metagenomics.
- Roland Krause
A short biography, summarizing Soren Brunaks kind introduction http://compbio.berkeley.edu/people...
- Roland Krause
The morphology of steves paper: http://www.improbable.com/airchiv...
- Shannon McWeeney
Intro: The ultraconservative (as seen from Berkely) and nonsense (as found in Through the Looking Glass
- Roland Krause
The jabberwocky poem does have meanings and is elegantly crafted.
- Roland Krause
Generally, nonsens in biology is bad.
- Roland Krause
Nonsense is generally bad, even in a codon
- Venkata P. Satagopam
Truncated proteins might interfere with physiological function (dominant negative). The cell removes such transcripts through nonsense-mediated decay (NMD).
- Roland Krause
Good example for NMD: Sox10
- Roland Krause
Mutations early in the gene leads to less severe phenotypes than later ones
- Roland Krause
NMD is an mRNA surveillance system
- Venkata P. Satagopam
NMD important to development of the immunesystem and cleans up other transcriptional errors.
- Roland Krause
We do not know how NMD works outside the mammals.
- Roland Krause
The mechanism involves the splicing machinery. If a stop is found wwithin 50nt upstream of the exon junction complex, it is removed..
- Roland Krause
50 nucleotide rule - translated normally or degraded by NMD
- Venkata P. Satagopam
brilliant nytimes article title - surviving on low number of genes
- Shannon McWeeney
splicing can introduce PTC - premature termination codon
- Venkata P. Satagopam
AS as mechanism to introduce PTCs - can lead to unproductive splicing
- Shannon McWeeney
these isoforms often have PTC
- Venkata P. Satagopam
Humans have fewer genes but better genes, due to AS.
- John Greene
from fftogo
Are PTC splice forms funcitonal?
- Venkata P. Satagopam
Many PTC mRNAs are noise
- Venkata P. Satagopam
analgous mechanism to shrinter: http://www.geeky-gadgets.com/cool-ga...
- Shannon McWeeney
Humans have fewer genes but better genes, due to AS.
- John Greene
from fftogo
Related reference: http://compbio.berkeley.edu/people...
- Shannon McWeeney
Alt splicing can yield isoforms differentially subjected to NMD
- Venkata P. Satagopam
SR protein - 11 in human which are serine and arginine rich
- Venkata P. Satagopam
SR proteins have premature stop codons.
- Roland Krause
SR genes has mRNAs with premature termination codons
- Venkata P. Satagopam
AS of PTC isoforms is mechanism for autoregulation of proteins
- Ted Laderas
NMD has a large effect on isoform abundance
- Venkata P. Satagopam
NMD has impact on isoform abundance - example of NMD clearing the major isoform
- Shannon McWeeney
minor isoforms are only shared 25% of time - modrek and lee 2003
- Shannon McWeeney
Not just anecdotal stories, splice patterns are conserved in mouse, implying functional significance.
- Roland Krause
(Unpulbished work)
- Roland Krause
All the SR proteins are talking to each other
- Venkata P. Satagopam
SR proteins 'compensate' for each other via coupling via AS and NMD
- Ted Laderas
SR genes have ultraconversed elements .. Bejerano et al 2004 Science 304: 1321
- Venkata P. Satagopam
Most ultraconserved regions are in intergenic regions, the regions in SR within genes.
- Roland Krause
question of why conserved - not protein coding, no obvious significant RNA secondary structure
- Shannon McWeeney
No proteins are produced from these genes.
- Roland Krause
The reason why SR sequences are highly conserved - most of the seq are not protein coding,
- Venkata P. Satagopam
no repetitive elements
- Venkata P. Satagopam
why conserved part 2 - no overrepresentation of binding / regulatory elements
- Shannon McWeeney
No simple explanations e.g. from miRNA binding etc.
- Roland Krause
no similarity elsewhere in genome except retropseudogenes
- Venkata P. Satagopam
analysis on origin of unproductive splicing
- Shannon McWeeney
No sequence similarity between the conserved elements. Seems to have been introduced mutliple times.
- Roland Krause
mouse and human SRp55 conserved but changing
- Venkata P. Satagopam
working on chordate SR proteins
- Venkata P. Satagopam
here intron and exon structure is more informative
- Venkata P. Satagopam
at this point - he has requested no further blogging - unpublished work
- Shannon McWeeney
no blog slides may be over
- Burkhard Rost
# Looks like interesting work.
- Roland Krause
wonderful talk
- Shannon McWeeney
Tells a (hard to blog) story about the successful treatment of collaborator with novel treatment based on genotyping.
- Roland Krause
Wow - what a conclusion! Fantatic talk...
- John Greene
from fftogo
# Certainly great work. The talk was nice too, and he only bitched at other reseachers in person once, another step up.
- Roland Krause
"Ultraconversed elements in SR genes ONLY show similarity to retropseudogenes" - what does this mean? Any takers?
- Saravanamuttu Gnaneshan
HL22: Dannie Durand - Sequence Similarity Network Reveals Common Ancestry of Multidomain Proteins
Ruchira S. Datta, Lars Juhl Jensen, Jason Stajich and arne liked this
Multidomain proteins are difficult to categorize because different parts have different histories.
- Gabriele Sales
Multidomain homolgs - finding homologs is an important aspect of functional genomics.
- Roland Krause
Song et al 2008 PLoS Computational Biology 4(5)
- Allyson Lister
Genes that share common ancestry tend to have similar structure and function.
- Gabriele Sales
also use to build comparative map of synteny
- Ruchira S. Datta
Sequence comparison can be used to identify chromosomal regions that share common ancestry.
- Gabriele Sales
this is called "spatial genomics"
- Ruchira S. Datta
How multidomain proteins fit this picture?
- Gabriele Sales
Example tyrosine kinases associated with many different domains.
- Roland Krause
example: protein tyrosine kinases
- Ruchira S. Datta
one family with many domain architectures, all sharing a kinase domain
- Ruchira S. Datta
Multidomain sequences evolve via gene duplication and domain shuffling.
- Gabriele Sales
multidomain sequences evolve via gene duplication and domain shuffling
- Ruchira S. Datta
The same domain may appear in multiple, unrelated proteins.
- Gabriele Sales
A definition will be presented that is in line with Fitch' proposition of homology.
- Roland Krause
can have case where genes share common ancestry, but domain architecture has changed
- Ruchira S. Datta
Difference between sequences related by vertical descent and related by domain insertion.
- Roland Krause
Two kinds of relations among genomes: relation by vertical descent or relation by domain insertion.
- Gabriele Sales
similarly can have the converse: through domain shuffling, genes that are not homologous can come to have the same domain architecture
- Ruchira S. Datta
It is possible to distinguish such two cases?
- Gabriele Sales
Given two sequences with similarity: Can one distinguish the two szenarios?
- Roland Krause
homologs are related by vertical descent
- Ruchira S. Datta
orthologs are related by speciation
- Ruchira S. Datta
orthologs are a subset of homologs, and homologs intersect with the set of significantly similar sequences
- Ruchira S. Datta
A Venn diagram, including orthologs, homologs, distant homologs and significantly similar sequences with modification.
- Roland Krause
also have distant homologs which don't appear to be significantly similar
- Ruchira S. Datta
inferences that can be drawn from vertical descent (similar molecular functions) and domain insertion (bindng partners) are different
- Allyson Lister
Biological interpretation of vertical descent: molecular function; regulation; comparative mapping; processes of gene duplication and genome rearrangement.
- Gabriele Sales
Interpretations of domain insertion: protein specialization; ligand specificity; localization; process of domain shuffling.
- Gabriele Sales
vertical descent implies similar: molecular function, regulation, comparative mapping, and is useful for processes of duplication and genome rearrangement
- Ruchira S. Datta
domain insertion leads to relationships of protein specialization, ligand binding, and cellular localization
- Ruchira S. Datta
In animals and plants multidomain sequences become more important than in bacteria.
- Gabriele Sales
The more higher eukaryotes will be sequenced, the more the problem needs to be addressed.
- Roland Krause
therefore, among similar sequences, want to distinguish which are related by vertical descent, and which by domain insertion
- Ruchira S. Datta
people look at sequence similarity E-value, and at alignment coverage
- Ruchira S. Datta
Alignment length is typically used to distinguish domain re-arrangements. Needs a decent mode model.
- Roland Krause
Good example that sequence similarity or e-values are not capable of distinguishing the two caes.
- Roland Krause
The goal of this method is to identify sequence pairs related by VD and DI,and should work on a broad range of families
- Allyson Lister
And needs to be computationally feasible.
- Roland Krause
To test, they looked at 20 well-studied families related by vertical descent.
- Allyson Lister
They had a much larger set of negative examples (40,000).
- Allyson Lister
PSI-BLAST performs worse then BLAST for sequences with variable architecture multi-domain proteins(!) as it pulls in non-homologous parts of sequences.
- Roland Krause
All methods do well with conserved multidomain proteins. They were more challenged by Variable multidomain, where Psi-BLAST doesn't do as well as BLAST. Both methods are extremely challenged when all the sequences were put into the analysis together.
- Allyson Lister
Pairwise comparisons are not sufficient. Try networks instead.
- Gabriele Sales
Pairwise sequences might not be enough, use the structure of the similarity networks.
- Roland Krause
Two sequences are compared in the context of their respective neighborhoods (i.e. other sequences that show similarity).
- Gabriele Sales
Domain architecture is implicitly present in the network.
- Allyson Lister
Open question. The model is explicitly based on insertion and deletion. What about de novo sequence formation?
- Gabriele Sales
Comment by Kevin Karplus: Use log scale for false positives in the ROC plots.
- Roland Krause
Keynote: Daphne Koller - Individual Genetic Variation: From Networks to Mechanisms
Colm Ryan, Andrew Su, Khader Shameer and 8 other people liked this
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
Few extra notes at http://www.fiamh.info/article...
- Oliver Hofmann
Other ways to read this feed:
