Harri Kiiveri

ICE: a statistical approach to identifying endmembers in hyperspectral images

Several of the more important endmember-finding algorithms for hyperspectral data are discussed and some of their shortcomings highlighted. A new algorithm - iterated constrained endmembers (ICE) - which attempts to address these shortcomings is introduced. An example of its use is given. There is also a discussion of the advantages and disadvantages of normalizing spectra before the application of ICE or other endmember-finding algorithms.

A general approach to simultaneous model fitting and variable elimination in response models for biological data with many more variables than observations.

BackgroundWith the advent of high throughput biotechnology data acquisition platforms such as micro arrays, SNP chips and mass spectrometers, data sets with many more variables than observations are now routinely being collected. Finding relationships between response variables of interest and variables in such data sets is an important problem akin to finding needles in a haystack. Whilst methods for a number of response types have been developed a general approach has been lacking.ResultsThe major contribution of this paper is to present a unified methodology which allows many common (statistical) response models to be fitted to such data sets. The class of models includes virtually any model with a linear predictor in it, for example (but not limited to), multiclass logistic regression (classification), generalised linear models (regression) and survival models. A fast algorithm for finding sparse well fitting models is presented. The ideas are illustrated on real data sets with numbers of variables ranging from thousands to millions. R code implementing the ideas is available for download.ConclusionThe method described in this paper enables existing work on response models when there are less variables than observations to be leveraged to the situation when there are many more variables than observations. It is a powerful approach to finding parsimonious models for such datasets. The method is capable of handling problems with millions of variables and a large variety of response types within the one framework. The method compares favourably to existing methods such as support vector machines and random forests, but has the advantage of not requiring separate variable selection steps. It is also works for data types which these methods were not designed to handle. The method usually produces very sparse models which make biological interpretation simpler and more focused.

Use of conditional probability networks for environmental monitoring

Causal or conditional probability networks (CPNs) are shown to provide a natural framework for combining a time sequence of classified satellite images with other maps for environmental monitoring. The key features of CPNs are described by way of application to an example involving the monitoring of salinization of farmland over time using satellite images and an ancillary dataset derived from a digital terrain model. It is shown that CPNs can be used to improve mapping accuracies by incorporating knowledge about the spatial and temporal variation of the map classes of interest. The methods provide a practical solution to the challenging problem of mapping and monitoring salt in farmland. The representation and propagation of uncertainty within this framework is discussed, as well as the spatial and temporal prediction of images and maps.

Image Fusion with Conditional Probability Networks for Monitoring the Salinization of Farmland

We show how a series of satellite images can be used in conjunction with data derived from a digital terrain model to monitor salinity in farmland. A conditional probability network (CPN) is constructed to produce salinity maps by combining uncertain information in images with uncertain knowledge or rules, where the rules are of a temporal and spatial nature. A specific model for extending conditional probability networks to handle the case of spatial context is given. To implement this model requires minor modifications to existing code for handling nonspatial CPNs.

High-throughput fabrication and screening improves gold nanoparticle chemiresistor sensor performance.

Chemiresistor sensor arrays are a promising technology to replace current laboratory-based analysis instrumentation, with the advantage of facile integration into portable, low-cost devices for in-field use. To increase the performance of chemiresistor sensor arrays a high-throughput fabrication and screening methodology was developed to assess different organothiol-functionalized gold nanoparticle chemiresistors. This high-throughput fabrication and testing methodology was implemented to screen a library consisting of 132 different organothiol compounds as capping agents for functionalized gold nanoparticle chemiresistor sensors. The methodology utilized an automated liquid handling workstation for the in situ functionalization of gold nanoparticle films and subsequent automated analyte testing of sensor arrays using a flow-injection analysis system. To test the methodology we focused on the discrimination and quantitation of benzene, toluene, ethylbenzene, p-xylene, and naphthalene (BTEXN) mixtures in water at low microgram per liter concentration levels. The high-throughput methodology identified a sensor array configuration consisting of a subset of organothiol-functionalized chemiresistors which in combination with random forests analysis was able to predict individual analyte concentrations with overall root-mean-square errors ranging between 8-17 μg/L for mixtures of BTEXN in water at the 100 μg/L concentration. The ability to use a simple sensor array system to quantitate BTEXN mixtures in water at the low μg/L concentration range has direct and significant implications to future environmental monitoring and reporting strategies. In addition, these results demonstrate the advantages of high-throughput screening to improve the performance of gold nanoparticle based chemiresistors for both new and existing applications.

Mapping and monitoring land use and condition change in the southwest of Western Australia using remote sensing and other data

In the south-west of Western Australia, the clearing of land for agricultural production has lead to rising saline ground water, resulting in the loss of previously productive land to salinity; damage to buildings, roads and other infrastructure; the decline in pockets of remnant vegetation and biodiversity; and the reduction in water quality. The region in question comprises some 24 million hectares of land. This has resulted in a wide variety of stakeholders requesting quantitative information regarding historical, present and future trends in land condition and use. Historically, two methods have been widely used to obtain information: (1) surveys requesting land managers to provide estimates of land use and condition; and (2) human interpretation of aerial photography. Data obtained from the first approach has in the past been incomplete, inaccurate and non-spatial. The second approach is relatively expensive and as a result is incomplete and is not regularly updated.In this paper, we describe an approach to land use/condition monitoring using remotely sensed and other data such as digital elevation models (DEMs). We outline our methodology and give examples of mapping and monitoring change in woody vegetation and salinity.

Multivariate analysis of microarray data: differential expression and differential connection

BackgroundTypical analysis of microarray data ignores the correlation between gene expression values. In this paper we present a model for microarray data which specifically allows for correlation between genes. As a result we combine gene network ideas with linear models and differential expression.ResultsWe use sparse inverse covariance matrices and their associated graphical representation to capture the notion of gene networks. An important issue in using these models is the identification of the pattern of zeroes in the inverse covariance matrix. The limitations of existing methods for doing this are discussed and we provide a workable solution for determining the zero pattern. We then consider a method for estimating the parameters in the inverse covariance matrix which is suitable for very high dimensional matrices. We also show how to construct multivariate tests of hypotheses. These overall multivariate tests can be broken down into two components, the first one being similar to tests for differential expression and the second involving the connections between genes.ConclusionThe methods in this paper enable the extraction of a wealth of information concerning the relationships between genes which can be conveniently represented in graphical form. Differentially expressed genes can be placed in the context of the gene network and places in the gene network where unusual or interesting patterns have emerged can be identified, leading to the formulation of hypotheses for future experimentation.

Fitting very large sparse Gaussian graphical models

In this paper we consider some methods for the maximum likelihood estimation of sparse Gaussian graphical (covariance selection) models when the number of variables is very large (tens of thousands or more). We present a procedure for determining the pattern of zeros in the model and we discuss the use of limited memory quasi-Newton algorithms and truncated Newton algorithms to fit the model by maximum likelihood. We present efficient ways of computing the gradients and likelihood function values for such models suitable for a desktop computer. For the truncated Newton method we also present an efficient way of computing the action of the Hessian matrix on an arbitrary vector which does not require the computation and storage of the Hessian matrix. The methods are illustrated and compared on simulated data and applied to a real microarray data set. The limited memory quasi-Newton method is recommended for practical use.

Environmental Monitoring Using a Time Series of Satellite Images and Other Spatial Data Sets

As a result of extensive farmland clearing over the last hundred years or so, dry-land salinity is a major problem in Western Australia. In fact, in some parts of the state, over 20 percent of Agricultural land is no longer productive. Prior to the work to be described in this chapter, no reliable large scale estimates of the extent or progression of salinity were available. This chapter describes a methodology for monitoring the historical extent of salinity, using a time series of satellite imagery, landform information derived from digital elevation models and ground truth data collected by experts with local knowledge. This work has served to highlight the salinity problem to decision makers in government and to provide input into the process of developing and applying remedial measures to arrest the spread of salinity.

PREDICTING ALZHEIMER'S DISEASE FROM A BLOOD-BASED BIOMARKER PROFILE: RESULTS FROM AIPL AT 54 MONTHS

O2-13-06 PREDICTING ALZHEIMER’S DISEASE FROM A BLOOD-BASED BIOMARKER PROFILE: RESULTS FROM AIBL AT 54 MONTHS Samantha C. Burnham, David Ames, David Baker, Ashley I. Bush, Lynne Cobiac, James Doecke, Kathryn A. Ellis, Noel Garry Faux, Richard Head, Harri Kiiveri, Simon Matthew Laws, Ralph Martins, Paul Maruff, Greg Savage, Bill Wilson, Lance Macaulay, Christopher Cleon Rowe, Colin Louis Masters, AIBL RESEARCH