What Causes Brain to Think?

Brain causes us to behave as who we are, but what causes brain to think (activate)?  Indirect measures of brain activities are usually preferred, such as fMRI, but the resulting statistics usually cannot resolve the causality issue.  The simple box plots (image below, left) show no significant differences of the brain activity under two experimental conditions.  Without modeling what the indirect response function looks like (e.g. HRF), our paper (to appear in Journal of the American Statistical Association) provides a way to resolve the causality question: what causes brain to think?(image below, right)

This method is specifically designed to provide causal interpretation under interference, and is robust against any shape of the response function.  The R package is CIN.

New R Package SCIO Posted

As promised, the new R package for SCIO (and post) has been uploaded to CRAN.  It will be publicly available across all platforms in a few days.

This is a version with the minimal set of functions for estimation.  I will add more functionalities in future releases.

I have also updated our R package clime.  Thanks useRs for all the suggestions.  The JASA paper is here, and the preprint is here.  I have also posted the cancer result using this method on this post.

Statisticians Compute Fast

Fast computation is increasingly important in analyzing data as we are living in an information explosion era.  Here is another paper we submitted and the preprint is on arXiv 1203:3896.  An example in the paper is to compute the brain networks of ADHD children and controls.  Here is a snap shot of the computation time.

The new method is called SCIO.  Will post the R package on CRAN.

Genetic Profiling for Breast Cancer Treatment

Our paper, CLIME, appeared in Journal of the American Statistical Association last year.  It is always amazing to me that how science can help shape the future.  Part of the paper employed a mathematical model to study if we can predict the outcome of a breast cancer patient based on her genetic profile.  The prediction performance is interestingly good.

Average (SD) performance of classifying breast cancer treatment outcomes over 100 out-of-sample runs.

Both the sensitivity and specificity are high (>=75%).  The R package for the CLIME method is here on CRAN, and I will provide an update very soon.