Whats meant by network learning ?
can we describe the data internally
can we describe the relationship between features as well
No just predicting outcome
Give some simple approaches for network learning
Covariance / correlation estimation
Inverse covariance matrix estimation
Name the rudimentary workflow for bio network learning
Get Data input
Process the data ( e.g. normalize it)
Learn a model on the data
Select the right model
Get a depency graph / covariance matrix
Name some real world networks
The internet
Sexual contact map
TARA ocean biology dataset
Food webs (biology, species, dependencies)
directed graph / who eats who
Name some networks in cellular biology
protein-protein interactions
Gene regulatory networks
Microbial networks
What are conceptual and computational challenges for network inference ?
data my be 0 inflated or compositional
datasets are undetermined p>>n
What is the MB method ?
sparse neighborhood selection
Find weighted graph by node wise linear regession
Use one OTU as Y and try to predict it
do this for every OTU
Link means that a node can be predicted by set nodes
we have to do model selection for each model
How can the problem of problem of spurious correlations be solved ?
suprious correlations ( correlations that apprear due to small sample size )
Example: David MacKays Gaussian quiz
covariance increases with distance (for X1)
conncetivity can be read from inverse covariance matrix (sparse)
Whats our key hypothesis for network inference ?
We expect a sparse network
Whats the GLASSO ?
-logdet prevents the determinate from being 0 and there fore losing a dim
S: sample correlation matrix
tr: sum of diagonal eletments
What are characteristics of bulk RNA seq ?
low number of samples
High dimensional
How can the validitiy of VST be assessed ?
plot rank(mean) vs. variance
if relationship is linear we need to normalize
plot rank(mean) vs. transformes variance
if the data is now normaly distributed ==> it worked
Do PCA on inital data
Do PCA on transformed data
If we can then seperate by condition / instead of any batch effect
Zuletzt geändertvor 5 Monaten
