Regression in connectomics?

[mrjiaruiwang]

We talked a lot about clustering and classification and statistical decision, but are there any applications for regression? I was thinking maybe trying to model phenomena with mathematical functions or even using regression to construct models for use in classification?

[akim1]

Calculating any of the coefficients in the models would be a form of regression analysis, no? Like 'p' in Bernoulli distribution?

[dlee138]

Using regression to construct models sounds like an interesting idea. By this, do you mean constructing a model graph from a known set of vertices and edges? For example, if you know a set of vertices and edges in a particular region of the brain and there's an unknown neuron in that region; you would be able to estimate what connections it has and the weights/likelihood of them based on known data. I'm not exactly sure how this problem can be solved, but perhaps we can map each known neuron in the brain to its geometric (x,y,z) coordinate and its connections, and run regression to figure out what possible connections an unknown neuron in the region could have.

[ghost]

For regression, the potential problems that I'm wondering about is given the size of data sets in connectomics, how computationally expensive is it to run regression? And also given that brains are somewhat non-linear, how useful is the model we make using regression?

[DSP137]

Regression analysis tries to determine relationships between predictor variables and a predicted variable, right? I think that was the goal of the MDMR for CWAS paper (trying to determine which voxel connections have a correlation with certain phenotypes). Perhaps a future goal for that work is to be able to determine which voxels are important for certain phenotypes and then use those voxels as predictors via some sort of regression to predict whether or not someone has said phenotype. I'm not sure how that would work in practice, though.