We introduce a framework for filtering features that employs the Hilbert-Schmidt Independence Criterion (HSIC) as a measure of dependence between the features and the labels. The key idea is that good features should maximise such dependence. Feature selection for various supervised learning problems (including classification and regression) is unified under this framework, and the solutions can be approximated using a backward-elimination algorithm. We demonstrate the usefulness of our method on both artificial and real world datasets.

No abstract

PBT2 is a copper/zinc ionophore that rapidly restores cognition in mouse models of Alzheimer's disease (AD). A recent Phase IIa double-blind, randomized, placebo-controlled trial found that the 250 mg dose of PBT2 was well-tolerated, significantly lowered cerebrospinal fluid (CSF) levels of amyloid-beta42, and significantly improved executive function on a Neuro-psychological Test Battery (NTB) within 12 weeks of treatment in patients with AD. In the post-hoc analysis reported here, the cognitive, blood marker, and CSF neurochemistry outcomes from the trial were subjected to further analysis. Ranking the responses to treatment after 12 weeks with placebo, PBT2 50 mg, and PBT2 250 mg revealed that the proportions of patients showing improvement on NTB Composite or Executive Factor z-scores were significantly greater in the PBT2 250 mg group than in the placebo group. Receiver-operator characteristic analyses revealed that the probability of an improver at any level coming from the PBT2 250 mg group was significantly greater, compared to placebo, for Composite z-scores (Area Under the Curve [AUC] =0.76, p=0.0007), Executive Factor z-scores (AUC =0.93, p=1.3 x 10(-9)), and near-significant for the ADAS-cog (AUC =0.72, p=0.056). There were no correlations between changes in CSF amyloid-beta or tau species and cognitive changes. These findings further encourage larger-scale testing of PBT2 for AD.

No abstract

Dementia is a global epidemic with Alzheimer's disease (AD) being the leading cause. Early identification of patients at risk of developing AD is now becoming an international priority. Neocortical Aβ (extracellular β-amyloid) burden (NAB), as assessed by positron emission tomography (PET), represents one such marker for early identification. These scans are expensive and are not widely available, thus, there is a need for cheaper and more widely accessible alternatives. Addressing this need, a blood biomarker-based signature having efficacy for the prediction of NAB and which can be easily adapted for population screening is described. Blood data (176 analytes measured in plasma) and Pittsburgh Compound B (PiB)-PET measurements from 273 participants from the Australian Imaging, Biomarkers and Lifestyle (AIBL) study were utilised. Univariate analysis was conducted to assess the difference of plasma measures between high and low NAB groups, and cross-validated machine-learning models were generated for predicting NAB. These models were applied to 817 non-imaged AIBL subjects and 82 subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI) for validation. Five analytes showed significant difference between subjects with high compared to low NAB. A machine-learning model (based on nine markers) achieved sensitivity and specificity of 80 and 82%, respectively, for predicting NAB. Validation using the ADNI cohort yielded similar results (sensitivity 79% and specificity 76%). These results show that a panel of blood-based biomarkers is able to accurately predict NAB, supporting the hypothesis for a relationship between a blood-based signature and Aβ accumulation, therefore, providing a platform for developing a population-based screen.