Grant: Discovery Grants Program - Individual
University of British Columbia
$25,000.00
Classification: methodology for variable selection and efficient tuning and comparasion of models
EF 2009-10
Résumé
Complex, high-dimensional data arise in many important classification problems. For example, in drug discovery, drug-like molecules are characterized by hundreds or thousands of chemical descriptor variables. The classification problem is to model the relationship between the chemical descriptors and the active/inactive assay response desired of the drug being developed, e.g., binding to a particular enzyme. Another data challenge is that active compounds are very rare in the assay data used to train models (the rare-class of interest problem).
As another example, in real-time genotyping for clinical intervention, multiple, redundant probes are used to compensate for the quick, and hence less reliable, chemistries employed. Here, the classification problem is to choose among the probe signal variables and determine which are most useful for genotyping each single nucleotide polymorphism (SNP), arguably the source of most genetic variation in humans, for a particular patient sample. Automated, real-time genotyping is an important step in personalized medicine. For clinical decisions, classification accuracy needs to be very close to 100%.
This proposal has three main components: (1) new modelling methods to improve classification accuracy; (2) methods for selecting among high-dimensional or redundant explanatory variables; and (3) computationally efficient cross-validation for assessment, tuning, and comparison of models or explanatory variable sets in terms of prediction accuracy. Although the scientific application focus is in drug discovery and genotyping, all the classification problems mentioned - rare class of interest, variable selection, model selection, and efficient cross validation - are pervasive in an extremely broad range of scientific disciplines.
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
British Columbia, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data
Tel que publié
Application id
36462-2008
Name nom
Welch, William
Department département
Statistics
Organization id
2
Institution établissement
University of British Columbia
Province en
British Columbia
Province fr
Colombie-Britannique
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2009
Competition year année de concours
2008
Award amount
25000
Program id
RGPIN
Program nane en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
Discovery grants programs
Group fr
Programmes de subventions à la découverte
Committee code
14
Committee name en
Statistical Sciences
Committee name fr
Statistique
Area of application code
504
Area of application group en
Manufacturing processes and products
Area of application group fr
Procédés et produits industriels
Area of application en
Human pharmaceuticals
Area of application fr
Produits pharmaceutiques administrables à l'homme
Research subject code
3000
Research subject group en
Statistics and probability
Research subject group fr
Statistique et probabilité
Research subject en
Statistics and probability
Research subject fr
Statistique et probabilité
Partie
2
Nb partie
5
Application title
Classification: methodology for variable selection and efficient tuning and comparasion of models
Keyword
NA
Application summary
Complex, high-dimensional data arise in many important classification problems. For example, in drug discovery, drug-like molecules are characterized by hundreds or thousands of chemical descriptor variables. The classification problem is to model the relationship between the chemical descriptors and the active/inactive assay response desired of the drug being developed, e.g., binding to a particular enzyme. Another data challenge is that active compounds are very rare in the assay data used to train models (the rare-class of interest problem).
As another example, in real-time genotyping for clinical intervention, multiple, redundant probes are used to compensate for the quick, and hence less reliable, chemistries employed. Here, the classification problem is to choose among the probe signal variables and determine which are most useful for genotyping each single nucleotide polymorphism (SNP), arguably the source of most genetic variation in humans, for a particular patient sample. Automated, real-time genotyping is an important step in personalized medicine. For clinical decisions, classification accuracy needs to be very close to 100%.
This proposal has three main components: (1) new modelling methods to improve classification accuracy; (2) methods for selecting among high-dimensional or redundant explanatory variables; and (3) computationally efficient cross-validation for assessment, tuning, and comparison of models or explanatory variable sets in terms of prediction accuracy. Although the scientific application focus is in drug discovery and genotyping, all the classification problems mentioned - rare class of interest, variable selection, model selection, and efficient cross validation - are pervasive in an extremely broad range of scientific disciplines.