Some links are currently broken … hang in there … working on updates
MIC2 R Package
The MIC2 R package implements Multilevel Integrative Clustering of brain cortical regions, imaged via electroencephalography (EEG). The goal of the package is that of producing statistical inference about patterns of synchronous brain activity, combining and borrowing strength from data collected longitudinally on multiple subjects. Given a longitudinal measure of electrode similarity, intended as coherence, cross correlation, etc., MIC formulates a hierarchical model for grouping electrodes on the cortex. The model jointly clusters electrodes over three levels of a hierarchy, so that a representative map of clustered cortical regions are available both at the level of specific individuals and as a group summary. The package includes, some data preprocessing routines, allowing for the estimation of similarity matrices longitudinally, given segmented multivariate time series. Graphical tools for visualization of results and inference are also implemented.
While the applications context is brain imaging, the technique can be easily generalized to accommodate various scientific contexts requiring the integration of data sources collected in a hierarchical fashion. For more of its details, please refer to our manuscript here (https://arxiv.org/abs/1609.09532).
This tool was created by Damla Senturk, Catherine Sugar, Donatello Telesca, and Qian Li, department of Biostatistics.
ucr R Package
ucr R Package
This R package implements joint curve registration and regression techniques for functional and longitudinal data. Functional variability is related to a set of predictors through regression models of functions amplitude (summarizing the overall strength of a functional signal) and phase (summarizing average timing of functional features). Supported models, include Gaussian, Poisson and Censored Gaussian sampling. Inference is based on Markov Chain Monte Carlo (MCMC) simulation. Details about modeling capabilities can be found in: Telesca et al. (2012), Erosheva et al. (2014), and Telesca (2015).
mombf R package
The mombf R package implements Bayesian model selection (BMS) and model averaging (BMA) for linear, asymmetric linear, median and quantile regression. This is the main package implementing the family of non-local prior (NLP) distributions (see Johnson and Rossell (2010, 2012) for a more detailed treatment), although other priors (mainly Zellner’s) are also implemented. The main features are:
Density, cumulative density, quantiles and random numbers for NLPs
BMS in linear regression (Section 1, Johnson and Rossell (2010, 2012).
BMA in linear regression (Section 4, Rossell and Telesca (2016).
Exact BMS and BMA under orthogonal and block-diagonal regression (Section 5, Papaspiliopoulos and Rossell (2016).
BMS and BMA for certain generalized linear models (Section 6, Johnson and Rossell (2012); Rossell et al. (2013)
BMS in linear regression with non-normal residuals (Rossell and Rubio, 2016).
Particular cases are Bayesian versions of asymmetric least squares, median and quantile regression. This manual introduces some basic notions underlying NLPs and illustrates the use of R functions implementing the main operations required for model selection and averaging. Most of these are internally implemented in C++ so, while they are not optimal in any sense they are designed to be minimally scalable to high dimensions (large p).
This tool was created by David Rossell, Donatello Telesca, and other contributing authors.
ToxTrees
ToxTrees
This software suite implements tree-based analysis of quantitative structure activity relationships associated with exposure or escalation experiments in toxicology and nanotoxicology. Given a library of chemical or nanomaterial stressors, their physicochemical characteristics are related to biological outcomes through regression trees with exposure response models on the tree leaves. Inference is based on Bayesian model averaging, providing full predictive distributions of exposure-response relationships. The outcome can be represented as dose-escalation only or escalation protocols involving dose and time of exposure. The software allows estimating variable importance, as well as marginal influence functions. Details about models and applications can be found in Low-Kam et. al (2015).