Welcome! My name is Timo Erkkilä. I'm a computer scientist working in life sciences developing scalable analysis tools for large, heterogeneous data sets. I work as a researcher at Tampere University of Technology (TUT), Tampere, Finland, in the group of Computational Systems Biology. I'm also a PhD student at Finnish Graduate School in Computational Sciences (FICS), Helsinki, Finland, and now I'm pursuing for a doctoral degree under the supervision of professors Harri Lähdesmäki and Olli Yli-Harja. I'm also affiliated to Schmulevich Group in Institute for Systems Biology, with whom I collaborate in The Cancer Genome Atlas.

Look me up on Facebook, Flickr, and LinkedIn.

One of my main focus areas is on developing computationally intensive statistical and machine learning tools for integration and analysis of genome-level data, and to provide visual aids in order to better grasp the regulatory processes governing complex biological systems and diseases. I'm involved in large international projects such as The Cancer Genome Atlas (TCGA) and Diabetes Prediction and Prevention (DIPP). In order to achieve the goals set in projects of this scope, substantial computational resources are required for making analysis tractable. Recent efforts include the use of Techila computing grid established at Tampere University Technology; the grid "borrows" unused computing capacity from hundreds of the campus area computers, transforming the network of idle computers into a real powerhouse. Techila was awarded the InnoFinland prize in 2009.

Various CSS style templates: here

RF-ACE

Regulome Explorer is a web application for visualization and exploration of multivariate statistical associations, inferred from a wide range of biological data. Currently the application is demonstrating how such diverse data as offered by The Cancer Genome Atlas can efficiently be explored:

Present-day systems biology studies often involve large quantities of mixed type data (categorical and numerical variables) with missing values. These characteristics severely limit the modeling possibilities one can consider when trying to uncover multivariate associations therein. The same applies to many other real world data analysis problems, where data sources are diverse and contain tens or even hundreds of thousands of candidate explanatory features. RF-ACE is an efficient feature selection algorithm best suited to aforementioned problems. It consists of a series of decision tree based feature selection procedures and statistical testing framework to accurately preform dimensionality reduction, identify statistically significant correlates, and predict data.

RF-ACE Google Code Project
RF-ACE Google Groups

mergedata is a generic program for merging data tables with row and column headers into a single data table. mergedata makes no assumptions on the order nor overlap of the data in the input data tables. However, rows and columns in the merged table appear in the input order:




Mergedata Google Code Project

DSection is a model for reconstructing cell type specific gene expression profiles from measurements of heterogeneous tissues. That is, no manual purification is needed, but the user must also provide information about cell type proportions in each tissue sample. Furthermore, samples can be categorized to experimental conditions -- the model then makes predictions on expression profiles across both cell types and experimental conditions.

Matlab scripts:
DSection.m (main program)
MCMCSummary.m
visualizeMCData.m

Check out my LinkedIn Profile for the most up-to-date Curriculum Vitae.

Erkkilä et al., "RF-ACE for Uncovering Nonlinear Associations from Heterogeneous Cancer Data", The Cancer Genome Atlas' 1st Annual Scientific Symposium, Maryland, USA, September 2011 [slides][talk]

Erkkilä et al., "Integrative Exploration of Cancer Data Reveals Regulatory Hotspots", WCSB, Zurich, Switzerland, June 2011 [slides]

Erkkilä et al., "Inferring genetic regulatory interactions from time-collapsed Boolean summary variables", WCSB, Luxembourg, June 2010


Erkkilä et al., "Probabilistic analysis of gene expression measurements from heterogeneous tissues", Bioinformatics, 2010 [article]

Dai et al., "A joint finite mixture model for clustering genes from independent Gaussian and beta distributed data", BMC Bioinformatics, 2009


Erkkilä et al., "A generic class density estimation framework using Dirichlet processes", WCSB, Aarhus, Denmark, 2009

Nikkilä et al., "Decomposing Gene Expression into Regulatory and Differential Parts with Bayesian Data Fusion", WCSB, Leipzig, 2008

Ruusuvuori et al., "Efficient automated method for image-based classification of microbial cells", ICPR, Tampa, Florida, USA, 2008