search

ataudt / chromstaR

Combinatorial and differential ChIP-seq analysis in R
8 stars 3 forks source link

Help in understanding the plot #17

Open sanchari24 opened 4 years ago

sanchari24 commented 4 years ago

Hello, I used chromstaR package in our lab. It was easy to use but I need a bit of your help to understand the computational aspect of it. I sincerely thank you for your time.

image

I have 5 histone marks and I ran the "Finding combinatorial chromatin states" analysis. I obtained the following plot attached below. Being a biologist, I am facing issues to interpret the plot. I have a time-point data for 0hr, 1hr and 6hr of stress in a plant specie and I feel this tool can give interesting result. The above plot is only for the zero hour (no stress).

1) What does "from" and "to" mean here biologically ? 2) Why is the heatmap asymmetric? Going from A to B and B to A, the probabilities are different ? 3) Why is the transition probability from going state A to state A not =1 ? 4) How to I export the values (transition probabilities) used to plot the heatmap ? 5) Which module should I use to generate the figure below for my data ? If I understand right, it talks about the presence of the classical marks and their combinations ?

image

ataudt commented 4 years ago

The first plot shows the transition probabilities in the Hidden Markov Model (HMM) that is built for the data. This model is a left-to-right model, which works well to predict peaks and chromatin states, but does not necessarily have a useful biological interpretation. 1) "from" and "to" here designate the probability to transition from a bin with combinatorial state "[from]" into a bin with combinatorial state "[to]". 2) The asymmetric probabilities are best explained with the empty state [ ] as example: Bins without any measured histone marks are the most frequent in the genome, so most other states transition into state [ ] with high probability. On the other hand, the empty state [ ] transitions into itself most of the time, since it is the most frequent state, thus the probability [ ]->[ ] is close to 1. 3) A transition probability of exactly 1 would mean that once the model is in that state, it would never get out of it. Even the probability [ ] -> [ ] is not exactly one, but slightly <1. 4) You can get the transition probabilities with model <- loadHmmsFromFiles("")[[1]] model$transitionProbs 5) The last subplot of the figure is described in section 7.3 of the vignette. The first and second subplot is custom code based on the data in model$bins$combination. The data for the first subplot (state distribution) should be retrievable with table(model$bins$combination). For the second subplot (state distribution in genes), you would first need to subset model$bins to only bins that overlap genes (using the GRanges package), and then do table(bins.subset$combination). Please also have a look at section 5.3 of the vignette. The figures in there might be even more suited for your needs.

Cheers, Aaron