@@ -7,7 +7,7 @@ Make sure you have the following Python packages installed:
* sk-learn
* MatplotLib
* Weblogo
* Bokeh
# Getting started
...
...
@@ -34,9 +34,16 @@ After creating the TFRecords, training set and test set, you can run 'train.py'.
# Interpreting the results
![Alt text](Interpretation.png?raw=true"Interactive visualization of what the network has learnt")
After evaluating the test set, the file interpretation.html will be created. In this file you will find two interactive plots:
* An interactive heatmap, which shows which filter pairs contributed to the classifcation of a sample
* A plot showing at which sequence position(s), of the sample selected in the heatmap, the filters found a match that was above the learnt similarity threshold
Select a sample and filter pair by clicking the corresponding tile in the heatmap. A tooltip with WebLogo's corresponding to a filter pair can be shown by hovering the cursor over a tile. A selected sample and filter pair will be plotted in the sequence position plot; that is, each position (horizontal axis) where a filter is found above the learnt detection threshold is annotated with a triangle (first filter in a pair) or an inverted triangle (second filter in a pair). The colour of a triangle indicates the similarity score between the filter and amino acids at the position the triangle is found.
# References
[1] Pan, X. Y., Zhang, Y. N., and Shen, H. B. (2010). Large-
scale prediction of human protein-protein interactions from amino acid
sequence based on latent topic features. Journal of Proteome Research,
