We also carried out some analysis to examine how well the ICA estimates could capture a certain relationship between ground truth and an Apoptosis Compound Library attribute of interest (phenotype). Toward this effort, we generated a set of pseudorandom vectors, each correlating with ground truth, with a preset correlation score (r = 0.5). Each vector

mimics a neurological, physiological, or physical attribute correlated with ground truth. We then computed how those vectors correlated with the ICA estimates. By generating multiple (N = 100) Inhibitors,research,lifescience,medical realizations of such vectors, and computing their correlations with the ICA estimates each time, we observed how accurately could the ICA estimates capture the relationship between “phenotypes” and ground truth. As the ICs from in vivo analysis do not have a ground truth, we sought to show that the select ICs indeed originated from metabolic sources, and not from confounds or nuisance artifacts Inhibitors,research,lifescience,medical associated with real data. To this end, we examined how the fractional tissue volumes in the spectroscopic voxel correlated with LCModel estimates or component weights. As cerebrospinal fluid (CSF) is mostly void of Inhibitors,research,lifescience,medical observable metabolites (Gasparovic et al. 2006), when the fractional tissue volume is high, more metabolites exist in the spectroscopic voxel and therefore both the estimates are

expected to correlate positively with the fractional tissue volume.

However, this effect is expected to disappear when the estimates are normalized with a reference metabolite estimate from within the voxel. The fractional tissue volumes in the spectroscopic voxel were estimated by segmenting high-resolution Inhibitors,research,lifescience,medical T1-weighted images into gray matter, white matter, and CSF using the unified segmentation approach available in SPM5 (Ashburner and Friston 2005) and averaging fractional tissue volumes of the T1-pixels within the spectroscopic voxel. Results The location of the spectroscopic voxel in vivo experiments, Inhibitors,research,lifescience,medical in the anterior cingulate region of the brain, is shown in Figure 1. Also shown is the LCModel output that presents a typical metabolite spectrum and LCModel’s fit to the spectrum; some key resonances are labeled and the estimated spectral baseline is also shown. Figure 1 Location of the voxel and typical MR spectra: 12 cc spectroscopic voxel and is positioned in the anterior cingulate region. LCModel plot of a typical in vivo metabolite spectrum in 1.8–4.2 ppm analysis window shows the real part of the referenced … Simulation As the composition of our simulated data is known, we only extracted as many ICs as the number of sources (12) underlying the data; ICs were paired with basis spectra, and corresponding component weights were also estimated.