We propose a method to jointly detect influential biomarkers and estimate how they change with dose. The assessment is made in dose-ranging trials collecting multiple outcomes for efficacy, safety, pharmacokinetics or pharmacodynamics. We regress all these outcomes versus a non-parametric transformation of the dose. The regression coefficients and the parameters from the dose-response model are simultaneously estimated using a penalized alternating least-squares method. We illustrate the technique with a phase I clinical trial and a metabonomic experiment in rats.