emission probability heatmap

[YichaoOU]

Hello,

How can I create the emission probability heatmap as shown in the ENCODE presentation?

https://wiki.uiowa.edu/download/attachments/101234675/Segway-UIowa-45-public.pdf?version=1&modificationDate=1375477976147&api=v2

[EricR86]

To create this figure, Segtools was used. Namely the tools segtools-gmtk-parameters. The figure is generated from the learned model parameters found after training inside the train directory (params/params.params).

Currently it's recommended to install segtools through Bioconda. Let me know if you need any further assistance.

[YichaoOU]

Thanks! Is row min =0 and row max=1, and they are emission probabilities?

[EricR86]

@YichaoOU the values across the row are normalized but they are not in themselves probabilities. For each row (track), this diagram shows which labels have a higher Gaussian mean distribution associated with each label. The black bars indicate the variance each underlying Gaussian has.

[YichaoOU]

Thanks! Do you know how I can directly plot emission probabilities? There are other plots from the segtools-gmtk-parameters command, but only the above figure looks like probabilities.

Is the table gmtk_parameters.stats.csv below the emission probabilities?

<style> </style>

	DNA_rep1	DNA_rep2	RNA_rep1	RNA_rep2
0	0.591853	0.605576	0.630044	0.645321
1	0	0	0	0
2	0.273725	0.298312	0.313079	0.348844
3	1	1	1	1

Thanks,
Yichao

[EricR86]

The parameters above are specifically just the learned Gaussian means from EM training.
If you want all the parameters to plot a Gaussian, the covariance needs to be obtained from the final params.params in your train directory. The covariance is tied across each track.

It is important to note that by default, all data input to Segway is normalized with an arcsinh transformation (there are options to change this). So the Gaussian means and covariance reflect these transformed values.

In params.params for the means (outside of the csv you posted above) you would look under the heading:
% means
And for the covariences under:
% diagonal covariance matrices