This computational methodology applies to measurements derived from mass spectrometry, or from chromatographic separations of biomolecular mixtures, for instance capillary or gel electrophoresis, affinity chromatography, or gel filtration of proteins or nucleic acids. These biomolecular mixtures normally originate from cell culture or tissue extracts, which may undergo prior purification to enrich a mixture with a single component, for instance phospho- or glycoproteins, nucleic acids containing certain sequences or bound to specific proteins. The mixture may also be subject to prior digestion of the components, e.g. treatment of proteins with proteolytic enzymes or DNA/RNA with restriction nucleases.

The methodology is based on recent developments in advanced machine learning techniques, involving classifiers, dimensionality reduction, and attribute selection. These techniques are coupled in an innovative sequence of steps that mathematically transforms the chromatography / mass spectrometry data into a form where it can be easily filtered to remove noise or unwanted systematic biases stemming from e.g. biological replicates of the experiment, or from use of slightly different protocols or instruments within an experiment. After the filtering, our method allows the mixture fractions specific for a certain physiological state to be distinguished more easily, for example the proteins expressed at higher levels in cancerous tissues than in healthy ones can be found with higher confidence.

Innovative Aspects:
The present invention provides several advantages over existing methods:
-it is based on a number of replicated experiments for each sample, drawing on statistical reliability
-it provides means for optimizing data representation parameters (e.g. resolution) by means of supervised machine learning algorithms
-it facilitates removal of components relating to noise and systematic errors in measurements.

This results in: (i) reliable determination of relevant fractions for the particular discrimination problem, and (ii) improved computational models for class distinction based on a filtered set of relevant fractions.

 

Patents/Rights: Patent(s) applied for but not yet granted

Type of Organisation: