Research

Next-Generation Sequencing (NGS), genetic polymorphisms (SNP, CNV), microRNA, Genome-Wide Association Studies, mass spectrometry, population genetics, feature selection, functional profiling

Next Generation Sequencing

Current projects. We develop Open Source modular pipelines for secondary and tertiary analysis of multi-platform NGS data sets, focusing on Illumina, SOLiD, Helicos. The pipelines integrate state-of-the-art software for data management, alignment and post-processing (e.g., bowtie, BWA, TopHat, SAMtools) with a predictive classification module based on mlpy for differential expression analysis (e.g., at gene, miRNA, or SNP levels). We recently developed a pipeline for phylogenetic profiling in metagenomics, with application to human gut NGS data.

Main collaborations. RIKEN Omics Science Center, Yokohama, Japan [P. Carninci, A. Forrest]. U.S. FDA and the Sequencing Quality Control project (SEQC) [L. Shi, K. Reagan]. University of Trento, CiBIO [A. Quattrone]. Istituto Tumori Genova [G.P. Tonini].

Genome-wide Association Studies

Single nucleotide polymorphisms (SNP) are genomic sites that differ by one nucleotide among a significant proportion (usually al least 1%) of a population or between paired chromosomes in an individual. SNPs are the main form of DNA sequence variation, occurring in a significant proportion of a population.

Current projects. We focus on the sources of variability in the analysis of GWAS data. In particular, we evaluate the effects of different batch sizes and case/control compositions on the outcome of GWAS results, using CHIAMO as genotyping algorithm on two Coronary Artery Disease datasets provided by the Wellcome Trust Case Control Consortium and the University of Ottawa Heart Institute.

Collaborations. U.S. Food and Drug Administration and the Genome Wide Analysis Working Group [F. Goodsaid].

Machine learning and Mass Spectrometry

Extracting predictive biomarkers from high-throughput mass-spectrometry (MS) data requires a complex analysis path, in which preprocessing and machine learning steps have to be pipelined to set up a predictive classifier, based on a shortlist of candidate features. The risk of overfitting is pervasive and it is thus important to carefully build a solid design analysis protocol (DAP).

Current projects. We developed an open-source Python/C software platform integrating both preprocessing and machine learning methods into a single package for the analysis of both proteomics and metabolomics MS data. MALDI-TOF and SELDI-TOF platforms are currently supported, while extension to LC-MS is under study. The preprocessing includes data I/O from/to different file formats, denoising, baseline correction, peak detection, peak alignment and extraction to a shortlist of features, which then undergoes a machine learning classification step performed by mlpy core.

As far as the preprocessing is concerned, our aim is to reduce the number of user-dependent parameters as much as possible; automatic tuning of preprocessing parameters is achieved by relating them to the technical parameters of the spectrometer (i.e., resolution and accuracy).

Collaborations. University of Trento, Edmund Mach Foundation - IASMA [P. Franceschi]

Publications

See http://mpba.fbk.eu/en/publications