Francesco Gadaleta, Ph.D.
Data Scientist
As a former member of the Statistical Genetics Unit at the Montefiore Institute, Department of Electrical Engineering and Computer Science of the University of Liège I have been involved in the FNRS Granted Integromics project to integrate heterogeneous omics data within a common framework.
My main research interests are penalised regression methods, bayesian statistics, nonlinear classifiers.
Since January 2015 I am part of a team, based at the Department of Human Genetics at the University Hospital UZ Leuven.
I am employed as Data Scientist operating in the healthcare sector, dealing with large datasets and custom machine learning algorithm which I follow from design to deployment.
High dimensionality of data and computational complexity require optimized solutions to deliver genetic data analyses in time.
I acquired the right skill set to improve and optimize large data systems in several domain (from healthcare to finance, social media, industrial, etc.)
The academic background provided me the most powerful theoretical tools for data analysis, such as neural networks and deeplearning, non-linear dynamics, Bayesian statistics, LASSO and regression, among others.
On the practical side I regularly use Linux, PHP, SQL, Amazon Web Service, R and Python.
As a member of a large team of researchers focused on the implementation and analysis of array CGH data in molecular diagnostics, I am specifically involved in a project related to the analysis of prenatal testing data.
Stuff I do
Here is a summary of what I have been busy with. This section is constantly updated to make room to new insights, projects and published work.
Integration of Gene Expression Data and Methylation Reveals Genetic Networks for Glioblastoma
Motivation: The consistent amount of different types of omics data requires novel methods of analysis and data integration. In this work we describe Regression2Net, a computational approach to analyse gene expression and methylation profiles via regression analysis and network-based techniques.
Results: We identified 284 and 447 unique candidate genes potentially associated to the Glioblastoma pathology from two networks inferred from mixed genetic datasets. In-depth biological analysis of these networks reveals genes that are related to energy metabolism, cell cycle control (AATF), immune system response and several types of cancer. Importantly, we observed significant over- representation of cancer related pathways including glioma especially in the methylation network. This confirms the strong link between methylation and glioblastomas. Potential glioma suppressor genes ACCN3 and ACCN4 linked to NBPF1 neuroblastoma breakpoint family have been identified in our expression network. Numerous ABC transporter genes (ABCA1, ABCB1) present in the expression network suggest drug resistance of glioblastoma tumors. Full pre-print on Arxiv.
Are we far from correctly inferring gene interaction networks with Lasso?
In this work we review nine penalised regression methods applied to microarray data to infer the topology of the network of interactions. We evaluate each method with respect to the complexity of biological data. We analyse the limitations of each of them in order to suggest a number of precautions that should be considered to make their predictions more significant and reliable. Full paper on Arxiv.
Network inference with Lasso (2013)
In the field of inferring Gene Regulatory Networks, I designed and implemented a penalised regression method that enhances the stability of discovered interactions by permutation test and scoring. Published on PLOS One.
MB-MDR. Just 100X faster (2014)
Model-Based Multifactor Dimensionality Reduction [Ref] in genetics has been designed to detect interactions of SNPs. I contributed to provide an approximated solution of the already faster version written in C++, that makes MDR just 100 times faster. Publication to appear. Binaries and user guide here.
PhD program at the University of Leuven
I conducted research in engineering the security aspects of hypervisors in the field of virtualization technology, the basic block of cloud computing. I designed and implemented countermeasures for native code vulnerabilities, runtime and virtualization environments.
I completed the PhD program with a dissertation titled “Virtualisation-based security countermeasures in software runtime systems” [free download]