Is it bad practice to change threshold for interaction probability?

Question

[gemmalouisebaldock]

Hi everyone,

I have 2 datasets that I will be using for my university project. The first dataset is relatively small and the other is much larger. Would it look odd in my university project to have a different threshold for the interaction probability for each of the data sets. The first one is smaller and I set the interaction probability to 0.5. Whereas the second dataset has several lists of proteins that are nearly 1000 proteins in length, so I will have to up the interaction probability threshold to 0.7 or 0.9 to make my Cytoscape/STRING network look less 'messy'. Is this bad practice?

Answer 1

[Carlo Yague]

While it is usually best and easier to explain/justify the use of a single threshold, I think it is ok to use different ones as long as it is properly explained. The true bad practice here would be to use different threshold and not make it clear. One way to be completely transparent about this would be to build 4 networks: the big and small dataset with both the high and low threshold. Then it is easy to discuss the benefits of each threshold by showing both of them and by explaining, for instance, why the low threshold is better for visualization of the small dataset and the high threshold is better for the visualization of the big dataset.