The development of DNA sequencing technologies allows us to collect large amounts of molecular data about the genome of each individual, and opens the possibility to precisely evaluate the risk of various diseases from one's molecular identity, or to rationally predict which drug is likely to be effective on a particular cancer. It also raises new challenges related to how to extract knowledge from large amounts of noisy data. In this context, I will discuss some regularization-based machine learning approaches we have developed to estimate complex, high-dimensional predictive models from relatively few samples, in particular in cancer prognosis and toxicogenetics.