Linking genotypes and phenotypes in personalized medicine

An appealing promise of personalized medicine is the use of molecular profiling technologies for "tailoring the right therapeutic strategy for the right person at the right time”. This implies, for instance, the use of genomic biomarkers to provide information about how a patient will respond to a given drug. This is, at the same time, an emerging need in several fields of clinical practice and a terrific challenge for biomedicine, genomics, statistics. To establish a quantitative link between genomic experiments and phenotypic evidence (e.g., response to a therapy), dedicated mathematical models are needed. While a number of such predictors have been used since a long time in other fields, they do not fit with genomic datasets. This is mainly because - differently from requirements of "classical" methods - the number of samples is by far smaller than the number of predictive genomic variables. I give a short review of the main problems affecting phenotype prediction from genotype data, and present a non-conventional statistical method developed at CRS4. I illustrate the results it provides when applied to an in vitro dataset of breast cancer cell lines subject to a series of non-disclosed therapeutic agents. While the method proposed has been applied to a particular biomedical task, it may serve the purpose whenever a quantitative link between genotype data and one or more phenotype features is desired.

@InProceedings{Mag14,
  author       = {Maggio, F.},
  title        = {Linking genotypes and phenotypes in personalized medicine},
  booktitle    = {Collana seminari interni 2014},
  number       = {20140318},
  month        = {march},
  year         = {2014},
  keywords     = {biomedicine, biotechnologies, genomica, molecular profiling technologies, non-conventional statistical method, personalized medicine, statistics},
  url          = {https://publications.crs4.it/pubdocs/2014/Mag14},
}