Presenter

Mr Michel Perez-Gonzalez - ETRO, Vrije Universiteit Brussel [Email]

Abstract

The increasing application of machine learning and artificial intelligence to a wide range of fields has opened new possibilities to analyze large amounts of data. Biotechnology has been one of the greatest beneficiaries of this strategy, for example adopting AI to streamline the development of new drugs.

Thanks to modern microscopes, it is currently possible to obtain vast amounts of information (sometimes in the order of millions of cells) at single-cell resolution level. This richness and abundance of information, in combination with sophisticated computer vision and machine learning techniques supported on data-intensive enabler platforms (cloud computing, HPC, GPU acceleration), have boosted considerably our ability to analyze and probe the cell as never before.

In this thesis we address the computational problem of automatic analysis of cellular dynamics from microscopic cell images, that take place in response to external stimuli (mechanical stimulation, exposure to drugs, and others). We cover the complete processing pipeline starting after image acquisition, which includes features extraction, cells segmentation, sub-cellular localization of proteins, until a methodology to summarize (with priorities) the most likely genes intervening in the cells adaptation response. We adopt an unsupervised approach for features extraction and cells segmentation to overcome the need for annotations in large population of cells. We contribute in these two tasks by (i) proposing methods that incorporate prior information about the cells, specifically about patterns expected to appear within osteoblasts and Yeast cell images. We approach the sub-cellular localization problem from a supervised multi-class classification perspective, and contribute with (ii) a novel features combination method (of linear complexity) that can learn involving millions of images with negligible computational overhead. For the dynamics extraction, we employ a network analysis strategy to identify the most relevant elements involved in the cells adaptation response.

Short CV

Master on Signals and Systems, Marta Abreu Central University of Las Villas, 2011