Prediction of long-range interactions in chromatin
Here at the Princess Margaret Cancer Centre, I have been working on designing and developing machine-learning methods to predict long-range interactions in chromatin, starting from DNase-Seq data and Hi-C data. My computational pipeline applies a deep learning algorithm.
Prediction of Gene Ontology (GO) annotations
During my PhD program, my main project has been to design and develop new machine-learning algorithms to predict Gene Ontology (GO) annotations. I departed from the common truncated singular value decomposition (tSVD) method, by first then developing some variants of it based upon clustering, and then trying some alternative techniques from topic modeling and deep learning. Among all the techniques I developed, the algorithm that led to the best results was a deep learning approach (precisely, a deep autoencoder neural network). This scientific research line finally produced two final main journal publications (BMC Bioinformatics and IEEE/ACM Transactions on Computational Biology and Bioinformatics, plus a third one currently under review on IEEE Transactions on Big Data), after many underway publications in conference proceedings and books.
The other project on which I have worked during my PhD program was called Search Computing. In this integrated web platform, I developed and integrated several web services able to query a genomics data warehouse, and to compute the semantic similarity between genes through the latent semantic indexing (LSI) method. This project led to a main journal publication (IEEE/ACM Transactions on Computational Biology and Bioinformatics) and to a technical report. The web services I developed are available on Bio Search Computing (Bio-SeCo) platform at http://www.bioinformatics.deib.polimi.it/bio-seco/seco/
See my last scientific articles on my publications page for further details.