KEDRI - Centre for Bioinformatics
Centre for Bioinformatics
This page is a summary of what our centre does and who we are. It covers:
The centre develops specific information processing methods and systems applicable to problems in bioinformatics and medical decision support.Research projects
- Computational modelling of gene/protein expression data and gene regulatory networks related to diagnosis and prognosis of disease and drug marker discovery
- Software system for gene expression analysis - SIFTWARE (Fig. 1)
- Software for gene regulatory network modelling
- Integrated systems for biological data analysis and modelling
- Nutrigenomic decision support systems for personalised advice
- Pest risk modelling and evaluation (with the CORE for Bioprotection, Lincoln University)
- Medical decision support systems. Personalised models for renal function evaluation (with Middelmore Hospital)
Figure 1. A snapshot of the patented Gene Profiling Software SIFTWARE developed with Pacific Edge Biotechnology Ltd (www.peblnz.com).
Personalized Modeling System for Renal Function Evaluation (GFR-TWNFI Homepage)
Project Team
- Ms. Tianmin Ma
- Dr. Mark Marshall
- Dr. Qun Song
- Prof. N. Kasabov
This project introduces the use of a novel transductive neuro-fuzzy inference model with weighted data normalization (TWNFI) for building personalized models for renal function evaluation. A software system that implements this application has been developed and made available to practitioners for experimental use.
While inductive approaches are concerned with the development of a model to approximate data in the whole problem space (induction), and consecutively – using this model to calculate the output value(s) for a new input vector (deduction), in transductive systems a local model is developed for every new input vector, based on number of closest to this vector data from the training data set.
In many neural network and fuzzy models and applications, raw (not normalized) data is used. This is appropriate when all the input variables are measured in the same units. Normalization, or standardization, is reasonable when the variables are in different units, or when the variance between them is substantial. However, a general normalization means that every variable is normalized in the same range, with the assumption that they all have the same importance for the output of the system. For many practical problems, variables have different importance and make different contribution to the output. Therefore, it is necessary to find an optimal normalization and assign proper importance factors to the variables.
The project describes the application of the TWNFI method for the task of personalized modeling for renal function estimation of patients and results are compared with other transductive or inductive methods. The TWNFI method not only results in a "personalized" model with a better accuracy of prediction for every single person, but also depicts the most significant input variables (features) for the model that may be used for a personalized medicine and improved treatment.
Computational Modelling of Gene Regulatory Networks GNetXP-“Gene Network Explorer” (GNetXP Homepage)
Project Team
- Prof. N. Kasabov - Coordinator
- Dr. Vishal Jain
- Dr. Dimiter Dimitrov, NCI, USA
- Dr. Igor Sidirov, NCI, USA
- Dr. Lesley Miller, Massey University
The main theme of the project is to model and understand behaviour of gene regulatory networks (GRN). Each gene interacts with many other genes in the cell, inhibiting or promoting, directly or indirectly, the expression level of messenger RNAs and thus the amounts of corresponding proteins. Transcription factors are an important class of regulating proteins, which bind to promoters of other genes to control their expression. Thus transcription factors and other proteins interact in a manner that is very important for determination of cell function. A major problem is to infer an accurate model for such interactions between important genes in the cell. To predict the models of gene regulatory networks it is important to identify the relevant genes. The abundant gene expression microarray data can be analysed by clustering procedures to extract and model these regulatory networks.
We have developed a novel method for GRN discovery of a large number of genes from multiple time series gene expression observations over small and irregular time intervals. The method integrates a genetic algorithm (GA) to select a small number of genes and a Kalman filter to derive the GRN of these genes. After GRNs of smaller number of genes are obtained, these GRNs may be integrated in order to create the GRN of a larger group of genes of interest. The method is designed to deal effectively with irregular and scarce data collected from a large number of variables (genes). GRNs are modelled as discrete-time approximations of first-order differential equations and Kalman Filter is applied to estimate the true gene trajectories from the irregular observations and to evaluate the likelihood of the GRN models. GA is applied to search for smaller subset of genes that are probable in forming GRN using the model likelihood as an optimization objective.
Siftware - Gene Expression Profiling Software System (Siftware Homepage)
Project Team
- Mr. Raphael Hu - Coordinator
- Prof. N. Kasabov
Gene expression data is characteristically highly dimensional with few samples. This can create great difficulties in selecting important genes from microarray data and creating models without using thousands of redundant variables. Siftware, which has been developed in conjunction with Pacific Edge Biotechnology Ltd (PEBL), is a software package which has been designed specifically for the purpose of analysing gene expression data.
Besides traditional modelling techniques such as linear regression and neural networks, Siftware includes also Evolving Connectionist Systems (ECOS) for classification tasks. This allows a level of profiling not possible in other modelling techniques. The ECOS model implemented here is the Evolving Classification Function (ECF). It allows for fast incremental learning and dynamic allocation of rule nodes. Rule nodes allow a level of transparency which allows biological experts to understand and therefore verify the model functionality. Siftware also contains a collection of more commonly used tools which can be split into three main categories. Firstly, for visualisation, Siftware uses 3-D visualisation and Principal Components Analysis (PCA). This is complemented by several clustering methods (hierarchical and k-means) for qualitative data analysis. Secondly, there are several feature selection methods including Signal to Noise Ratio (SNR) and the Student’s t test. Finally, for comparative testing, Siftware also incorporates Multiple Linear Regression (MLR).
Nutrigenomics and Biomedical Ontologies
Project Team
- Prof. N. Kasabov
- Dr. Anju Verma
- Mr. Paulo Gottgtroy
This project is focusing on nutrigenomics related to aging and diabetes, with the aims of utilizing genomic data for personalized dietary advice, and managing and organizing metadata related to nutrigenomics ontologies related to diabetes and aging. Microarray data from experiments of diabetic vs healthy and old vs young patients is linked with nutritional data and artificial intelligence methods used to pinpoint genes of interest and diet components of relevance for healthy and disease-preventing advice.
Chronic Disease Ontology (Download)
Pest risk evaluation and modelling
Project Team
- Prof. N. Kasabov
- Dr. Snjezana Soltic
- Dr Andrew Lowe, Queensland University
- Dr. Susan Worner, Lincoln University
- Dr. Paul Pang
Plant and insect pests are of great importance to New Zealand and Australia, and predicting and modelling the risk of different pest species before and after invasion is crucial for their efficient management and mitigation of damage to agriculture and environment. This project is looking into neural network modelling as a tool to predict and build statistical models of pest invasion and movements, based on climate, host range, pest distribution and other variables. Evolving connectionist systems are applied to develop personalized predictions of pest movement and outbreak probabilities for practical countermeasures to be taken.
DOPPS-TWNFC: Evolving Systems for Medical Prediction
Hemodialysis has been one of most striking successes of modern healthcare. There are an estimated 1.15 million patients worldwide treated with dialysis. Accurate prediction of hemodialysis patients’ longevities are needed. A system that allows for accurate prediction of patients’ survival outcomes will potentially allow dialysis providers to achieve equitable and improved outcomes to a heterogenous group of patients on dialysis. Such system will also potentially allow funding agencies to anticipate and absorb increasing dialysis costs, and allow accurate decision-making by the dialysis provider, patient and family to withhold dialysis for those whose survival and/or quality of life is going to be very poor.
Data originate from the Dialysis Outcomes and Practice Patterns Study (DOPPS). The DOPPS is based upon the prospective collection of observational longitudinal data from a stratified random sample of haemodialysis patients from the United Sates, 8 European countries (United Kingdom, France, Germany, Italy, Spain, Belgium, Netherlands, and Sweden), Japan, Australia and New Zealand.
TWNFC is a dynamic neural-fuzzy inference system with a local generalization, in which, the Zadeh-Mamdani type fuzzy inference engine is used. To partition the input space for creating fuzzy rules and obtaining initial values of fuzzy rules, the Evolving Clustering Method (ECM) is applied and the cluster centers and cluster radiuses are respectively taken as initial values of the centers and widths of the Gaussian membership functions. A steepest descent back- propagation (BP) learning algorithm is used for optimizing the parameters of the fuzzy membership functions. The systems developed in this project performs a better local generalization over new data as it develops individual models for each data vector that takes the location of new input vector in the space into account. The systems developed in this project will provide immediate clinical tools to augment renal health care delivery that are workable and of sufficient accuracy to support good clinical decision making. A software system that implements this application is being developed for experimental use by practitioners.
Our staff
- Prof. Nikola Kasabov, Centre Director
- Dr. Dave Parry
- Dr. Colleen Higgins, Associate Director
- Dr Yingjie (Raphael) Hu
- Linda Liang
- Shoba Tegginmath
- Michelle Jamieson
External collaborators
- Prof. Dimiter S Dimitrov - NIH (USA)
- Prof. Vladimir Brusic - Dana-Farber Cancer Institute, Harvard Medical School (USA)
- Prof. Allen Rodrigo - University of Auckland
- Prof. Tony Reeve - University of Otago
