Machine Learning for Personalized Medicine

Project: Analysis and detection of differential isoform expression and splicing dysregulation events from RNA-seq data

Researcher: Max Zwiessele	Supervisor: Neil Lawrence	University of Sheffield Sheffield, UK

Max Zwiessele was born in Esslingen, Germany, in 1988. He studied Bioinformatics at the University of Tübingen, and obtained his MSc in 2012 working on probabilistic modelling of Microbiological data using Gaussian processes.

Project description

I am working on a probabilistic model, called Baysian GPLVM (Gaussian Process Latent Variable Model). It is a probabilistic interpretation of PCA (Principal Component Analysis), allowing for non-linear mappings for regression or dimensionality reduction. Probabilistic modelling of genomic traits allow for inclusion of different sources of data, which will help to improve predictive performance of mapping molecular traits to observed phenotypes.

My goal is to develop a unifying probabilistic model for continuous and discrete data, which is able to include structural and hierarchical information about biology. I am using Bayesian GPLVM to accomplish this task, because of its unique features discovered in the past. It is capable of taking advantage of structural information such as time dependency (Damianou et al.) or taking unknown environmental perturbations into account (Fusi et al.).

This model became feasible as new discoveries allow for ‘big data’ using stochastic learning (Hensman et al., stochastic gradient descent) and distributed computation (Gal et al.). 

Code:
All my implementations are publicly available on Github in the Gaussian process toolbox GPy (Github). I am a developer on the framework, implementing dimensionality reduction tools using probabilistic models. 

Other Projects:
I am organizer in the Gaussian process summer school held every half year in the University of Sheffield (GPSS). If you want to learn more about Gaussian processes or are a beginner and want to start using GPs feel free to participate. It is a great opportunity and will help you understand Gaussian processes more.

Motivation for participating in the network

Personalized medicine is becoming a big part in the healthcare of humanity. Personalized medical care tailored to each individual will bring a more healthy future with higher accuracy in curation of all ailmends and less drug usage.

               ********************************

Duration of fellowship: April 2013 to April 2016

Contact: ibinbei@gmail.com

Homepage: http://staffwww.dcs.sheffield.ac.uk/people/M.Zwiessele/

Publications:

ARTICLES
- James Hensman, Max Zwiessele, Neil Lawrence
. Tilted Variational Bayes. JMLR W&CP 33 : 356–364, 2014.

- Ricardo Andrade Pacheco, James Hensman, Max Zwiessele, Neil Lawrence
. Hybrid Discriminative-Generative Approach with Gaussian Processes. JMLR W&CP 33 : 47–56, 2014.

CODE
General implementations
You can find almost all of my code in the GPy package which can be found on Github (GPy).

IPython notebooks
For some IPython notebook examples and more explanation of the toolbox see GPy Notebooks. 

How-to view IPython notebooks
You can run the notebooks by visiting nbviewer and giving the address of the notebook in Github. For example try 

Coregionalized Regression Model