University Logo
Sydney Bioinformatics and Biometrics

The projects that the group members work on can be placed into seven main categories.

Multi-omics Integration and Statistical Machine Learning

Biological systems are complex and the observed variablity between samples may be explained by different kinds of 'omic' information, such as the transcriptome or the proteome. However, different omics varieties exhibit unique characteristics and require tailored approaches to draw meaning from these datasets by addressing statistical issues such as sample mismatch, missing values, imbalance of features and differing requirements regarding normalisation or standardisation. Our research focuses on building better predictive models for biomarker discovery, which involves developing methods to examine clinical-pathological, mutation and omics data for prognostic biomarker discover, for example constructing suitable weights to improve weighted lasso.

Representative Publications

Jayawardana, K., Schramm ,S.J., Haydu, L., Thompson, J.F., Scolyer, R.A., Mann G.J., Müller S. and Yang J.Y. (2015) Determination of prognosis in metastatic melanoma through integration of clinico-pathologic, mutation, mRNA, microRNA, and protein information. International Journal of Cancer. 136(4):863-874.

Strbenac, D., Mann, G., Ormerod, J. and Yang, J. (2015) ClassifyR: an R package for performance assessment of classification with applications to transcriptomics. Bioinformatics. 31(11):1851-1853.

Patrick, E., Buckley, M., Müller, S., Lin, D. and Yang, J. (2015) Inferring data-specific micro-RNA function through the joint ranking of micro-RNA and pathways from matched micro-RNA and gene expression data. Bioinformatics. 31(17):2822-2828.

Available Projects

There are currently 3 positions open to applicants.

Sample-wise error rates for individual and combined datasets

Complex Data and Network Modelling

Biological systems are made up of a highly connected set of features and often feature built-in redundancy, so that damage to one component may be unimportant if another component which can perform the same function is undamaged. Considering biological molecules in context of their upstream regulators and downstream targets, as well as alternative pathways that lead to the same target allows better association between measured biological changes and observed phenotypes. Statistical models need to incorprate uncertainty into the inference procedure because network models are obtained from publicly available databases derived from datasets comprised of different organisms and different cell types than the ones being studied.

Representative Publications

Ghazanfar, S. and Yang, J. Y. (2016) Characterizing mutation-expression network relationships in multiple cancers. Computational Biology and Chemistry. 63:73-82.

Wang, R., Jiang, K., Feldman, L. J., Bickel, P. J. and Huang, H. (2015) Inferring gene-gene interactions and functional modules using sparse canonical correlation analysis. Annals of Applied Statistics. 9(1):300-323.

Barter, R., Schramm, S., Mann, G. and Yang, J. (2014) Network-based biomarkers enhance classical approaches to prognostic gene expression signatures. BMC Systems Biology. 8(Supplement 4):S5.

Available Projects

There are currently 2 positions open to applicants.

Coloured Graph

Precision Medicine

Methods are being developed to bring together information from health or phenotype data together with omics information to provide better predictions of patient responses and outcomes. Apart from data integration, the understanding of signalling, proteomic and transcriptional regulation is a focus of research, in particular for stem cell differentiation and early development. Such understanding is crucial for precision medicine as it builds the foundation for testing drugs, treating or preventing birth defects, as well as other complex diseases caused by the malfunction of the complex regulatory networks.

Representative Publications

Patrick, E., Schramm, S.-J., Ormerod, J.T., Scolyer, R.A., Mann, G. J., Müller, S. and Yang J. Y. (2017) A multi-step classifier addressing cohort heterogeneity improves performance of prognostic biomarkers in three cancer types. Oncotarget. 8:2807-2815.

Available Projects

There is currently 1 position open to applicants.

DNA Stethoscope

Model Selection and Visualisation for Biomedical Data

Many classification tasks with high-throughput biological data have high error rates, suggesting multiple distinct features classify different subsets of individuals or that simple changes do not capture the full set of differences present among classes (e.g. recovery vs. cancer relapse). Motivated by complex data, including for longitudinal, clustered and correlated univariate and multivariate responses, classification and improved prediction methods are developed for multi-omics data. Theoretical properties of regularization methods in various asymptotic scenarios are investigated and improvements to these methods have been made by learning from resampling-based stability information. Advanced model visualisation methods are implemented to enable interactive and dynamic model building allowing for the investigation of robust selection and estimation methods.

Representative Publications

Tarr, G., Müller, S. and Welsh, A. (2017) mplot: An R package for graphical model stability and variable selection. Journal of Statistical Software. To appear.

Strbenac, D., Mann, G. J., Yang, J. Y. and Ormerod, J.T. (2017) Differential distribution improves gene selection stability and has competitive classification performance for patient survival. Nucleic Acids Research. 13:e119.

Available Projects

There is currently 1 position open to applicants.

Differential Distribution Classifier

Single Cell Data Analytics

Until recently, heterogeneity across single cells in a biological tissue extract had been lost when a sample comprising thousands of cells was measured on the same assay, providing only an average measurement and thus discarding information about cell-to-cell variation. Our research focuses on analysis of single cell RNA-Seq (scRNA-Seq) data, but extends to other technology, such as CyTOF. We have employed statistical mixture modelling to reduce the impact of noisy measurements and to characterise the expression profiles of genes across populations of single cells. We have used this framework to characterise the expression profiles of single neurons using a variety of publicly available scRNA-Seq datasets.

Representative Publications

Ghazanfar, S., Bisognni, A. J., Ormerod, J.T., Lin, D. M. and Yang, J. Y. (2016) Integrated single cell data analysis reveals cell specific networks and novel coactivation markers. BMC Systems Biology. 10(Supplement 5):127

Single Cell Expression of Gene

Brain Omics and the Connectome

Recently, the capacity to analyse the large complex datasets generated in studies of brain function and related disorders has begun being implemented. This benefits from the team's strengths in hypothesis formulation, network construction, image analysis and data integration to extract information from various high-throughput technologies such as functional magnetic resonance imaging (fMRI) and microscopy.

Representative Publications

Patrick E., Rajagopal, S., Wong, H-K. A., McCabe C., Xu, J., Tang, A., Imboywa, S.H., Schneider, J.A., Pochet, N., Krichevsky, A.M., Chibnik, L.B., Bennett, D.A. and de Jager, P .L. (2017) Dissecting the role of non-coding RNAs in the accumulation of amyloid and tau neuropathologies in Alzheimer's disease. Molecular Neurodegeneration. 12:51.

Available Projects

There is currently 3 positions open to applicants.

Artistic Network

Approximate Bayesian Inference

Methods in this category are sometimes necessary for large, complex problems. By exploiting the flexibility of Bayesian hierarchical modelling framework, these models can be fit using variational Bayes, expectation propagation and Laplace-like approximations in mere minutes whereas exact methods, such as Markov chain Monte Carlo (MCMC), could take days to produce a result.


Complex Agricultural Data Modelling

Animal and plant breeding trials are routinely conducted with the aim of improving food quality and quantity via selection of superior breeds. With the advent of both high-throughput genotyping and phenotyping technologies, there is now an abundance of data. The optimal way to design and collect these data and the inference of meaningful interpretation from these big datasets is the focus of research efforts. Specific examples include construction of genetic linkage maps, quantitative trait loci analysis, factor analytic mixed model analysis of crop breeding trials. Strong links with Meat and Livestock Australia have also been established.

Representative Publications

Konarska, E., Kuchida, K., Tarr, G. and Polkinghorne, R.J. (2017). Relationships between marbling measures across principal muscles. Meat Science. 123:67-78.

Norman, A., Taylor, J, Tanaka, E., Telfer, P., Edwards, E., Martinant, J.P. and Kuchel, H. (2017) Increased genomic prediction accuracy in wheat breeding using a large Australian panel. Theoretical and Applied Genetics. 130:2543-2555.

Arbitrary Densities