Amélioration du contrôle de qualité de produits sanguins utilisant la spectrométrie de masse à haut-débit et l'apprentissage automatique

Authors: Brochu, Francis
Advisor: Corbeil, JacquesLaviolette, François
Abstract: This memoir describes work concerning the treatment and analysis of high-throughput mass spectrometry. Mass spectrometry is a tried and tested method of chemical measurement in a sample. Applied to biological samples, mass spectrometry becomes a metabolomic measurement technique, meaning that it measures the metabolites contained in a sample, which are small molecules present in the biological fluid that interact with the individual’s metabolism. The project that is presented here is a partnership with Hema-Québec in order to conceive new quality control tests from mass spectrometry measurements. The application of the LDTD ionisation source in mass spectrometry makes the acquisition of spectra in high-throughput possible. This represents a large benefit in terms of experimental costs and in time. Large datasets of mass spectra can then be obtained in a short period of time. The computer science domain of machine learning can then be applied to this data. Statistical machine learning can then be used to classify the spectra of blood product samples and provide statistical guarantees on this classification. The use of sparse and interpretable machine learning algorithms can also lead to the discovery of biomarkers. The work presented in this memoir concerns the design of two methods of treatment of mass spectra. The first of these methods is the correction by virtual lock masses, used to correct any uniform shift in the masses in a spectra. The second is a new method of peak alignment used to correct slight measuring errors. In addition, a new kernel method, a method to mathematically compare examples, was designed specifically for application on mass spectra data. Finally, results of classification on mass spectra acquired with an LDTD ionisation source and by liquid chromatography mass spectrometry will be presented.
Document Type: Mémoire de maîtrise
Issue Date: 2018
Open Access Date: 30 May 2018
Permalink: http://hdl.handle.net/20.500.11794/29875
Grantor: Université Laval
Collection:Thèses et mémoires

Files in this item:
SizeFormat 
33988.pdf1.16 MBAdobe PDFView/Open
All documents in CorpusUL are protected by Copyright Act of Canada.