• Model selection annd variable selection
• Multi-state models and survival analysis
• Multiple testing problems,
• Dimension reduction methods
• Analysis of high dimensional and massive data set such as “Omics” data 
• Bayesian Modelling
• Machine Learning
• Application to  combination of clinical, health and biological data
• Statistics methods to Computational Ecology and Environmental Sciences

• STAT7111/8111 (Generalized Linear Model), S2, 2020
• Mathematical Background for Biostatistics (MBB), S2, 2020 at BCA. 

• The Mathematical Engineering of Deep Learning  during the summer school organized by AMSI https://ss.amsi.org.au/

PhD scholarships

If you are interested in doing a PhD or MRes at Macquarie under my supervision, and are eligible to apply for any of the scholarships listed here, then please contact me in advance of the closing dates.

Postgraduate research topics in statistics:

Competing risks for clustered data: the heterogenity of the COVID mortality across countries

Supervisor: Benoit Liquet

Topic Description:

The mortality rate of COVID-19 among patients admitted to the intensive care unit (ICU) is currently very high. It is one of the primary cause of death in ICUs. The aim is to build a prediction model that predicts the risks and chances of the expected course of intensive care unit (ICU) admissions for suspected COVID infected patient. The model should provide evidence based information for physicians.

The specific aims are to develop and implement an updating mechanism which allows dynamic updating of the predictions for a given patient when important markers change, and dynamic updating of a model which was adapted to a given ICU when new data was entered into the database. To achieve it, we will exploit combined data sets which detail a variety of patients entering in ICU from diverse geographic locations. Then, the heterogeneity of practice across countries will be addressed.

The main aims of this project are to:

•          Develop a multi-state model which models the course of the patient in intensive care unit.

•          Different Markov models will be investigated: non-homogeneous Markov models; semi-markov Models and finally non-homogeneous semi-markov Models

•          An additional avenue is also to study the ``landmark’’ model

•          Incorporate random effect (“Frailty Models”) to take into account potential heterogeneity

•          Assessment of prediction accuracy

•          Development of R packages.

•          Writing of scientific article

Joint sparse group Bayesian models on individual data for detecting pleiotropic effects: application to cancer datasets

Supervisor: Benoit Liquet

Topic Description:

Recent technological advances in molecular biology and genomics given rise to numerous large-scale datasets. The sheer size and complexity of these data sets imposes new methodological challenges. The aim of this project is to boost and enhance the current toolkit for integrative analysis of massive datasets, especially for detecting pleiotropy.

The novel methodology that we aim to develop will enable scientists to explore and analyse massive data sets from the “dark genome”, majority of genes in the genome have minimal knowledge. This research contributes to extend pleiotropy knowledge in human disease in order to drive diagnosis, therapeutic intervention and individualised treatment within precision medicine.

The new statistical approaches will be applied to enrich our insights about the genetic mechanisms of thyroid and breast cancer types.

The main aims of this project are to:

•          Methods development for pleiotropy analyses

•          Simulation studies design and validation of methods.

•          Application to real data. Comparison of results while considering each cancer individually and while considering both cancers together.

•          Development of R packages.

•          Writing of scientific article

Advanced Machine Learning methods for Corals classification within the lagoon of Maupiti

Supervisor: Benoit Liquet

Partners: Damien Sous (MIO/SIAME), Samuel Meulé (CEREGE)

Topic Description:

Mapping Coral Reef Ecosystems are crucial for monitoring, protecting and ressource management purpose. Remote sensing by satellite offers a valuable cost-effective and time-efficient tool to provide image data of Coral reefs. A major challenge is to be able to extract information on coral classification and habitat description from the satellite data, in order to track the evolution of coral reef-lagoon systems in a global degradation context. A first accurate processing of the images is carried out to extract statistical metrics of the reflectance in different bands. These parameters are used to build an extensive database of explanatory variables. A series of classification tools can thus be applied to discriminate and to classify the different coral types. In particular, supervised classification methods could be explored such as:

•          Softmax regression model

•          Support Vector Machine

•          Random Forest

•          Gradient Boosting tree

In this project, different Machine learning methods for Corals classification will be compared in order to provide the best tool for pixel classification. A significant outcome will be the creation of an automated computational R package to classifies each pixel in an image given a pre-defined set of classes.