1

The image below shows a biological network where nodes have been colored based on a set of measurements. enter image description here

This works great, but the next step is to overlay up to eight sets simultaneously. This means that instead of having a single color for one node, there could be up to eight colors for each highlighted node.

I have tried more than a dozen versions so far but I am not quite sure I found the best solution yet.

The challenges so far have been:

  • the fixed layout of the network, nodes are manually laid out and cannot change position so any solution that changes the nodes shapes or sizes is not good
  • the network is already cluttered, so adding more colors and shapes can end creating a mess

The whole purpose of doing this multiple overlay is to make it easier for users to find patterns in the data, nodes that are similar across multiple sets of measurements.

I would share some of the best solutions I found but I am afraid I would lead or mislead you into the same pits I have falled myself and this whole excercise would be pointless.

1

Keep data perceivable

Visualizing eight dimensions on one small point seems like a lot to ask. Even if you found a solution that looks nice it will probably be outside your requirement of making it "easy to find patterns".

Is it possible to average the data sets? You could visualize the average and expand the details of a node on interaction.

There might also be value in allowing the user to select what data set is visualized. Then they could compare the results across all nodes for a discrete data set of their choice.

Manage density

The problem is that you already have several dimensions in play:

  1. Network connections (arrows)
  2. Node groups (surrounding shape)
  3. Node type (shape)
  4. Node description (text label)
  5. Measurements (color)

At a certain point (often around 3-4 dims, 5 if you're awesome) you reach diminishing returns. Visualizing all the things means patterns will be obscured not revealed.

What about a way to switch views for different types of analysis?

  • Network
  • Measurement
  • The other thing they want to know ...

This opens up a lot of potential to do more without overloading what the UI already enables.

| improve this answer | |
  • You are right, eight is a lot to ask. The good part is that is highly unlikely any user would find value in displaying all of them. It is likely that they would chose a combination of two, three or four. Averaging the dimensions for each node would not work since the measurements are quite different and each one has a story to tell. Right now, they can select which dimension to visualize but this is not enough to help them find patterns accross dimenisons. – Ovidiu Berdila Dec 10 '15 at 19:08
  • Thank you for updating the answer. I also think this is the best thing to do. I am already working on a separate view with parallel coordinates where nodes overlap and it's easier to spot the patterns. – Ovidiu Berdila Dec 11 '15 at 8:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.