The experimental characterization of the denatured state of proteins is still an open question, despite the heavy effort made in the last four decades to investigate it. The biologically-relevant disordered phase, in water, is not experimentally accessible, and little is known about the molecular mechanism of chemical-induced denaturation, and to which extent the chemical-induced denatured state is similar to the metastable denatured state in water. In the first part of the thesis, we address this problem by means of a computational approach. Using bias-exchange metadynamics simulations, we investigate the conformational properties of two peptides (the alpha-helix and a beta-hairpin of protein G B1 domain) in urea and guanidine hydrochloride solutions and we compare them with the denatured state simulated in pure water. We find that the denaturation mechanisms seem to be different from the two-state picture commonly assumed : the content of secondary structure is rather heterogeneous in the denatured alpha-helix, whereas the denatured beta-hairpin shows basically only beta-content, with different weights in the different solvents. Moreover, urea displays a local effect, establishing a competition with water in forming hydrogen bonds with the residues, while the guanidine hydrochloride affects the whole secondary structure via a long-range, electrical interaction. To perform the simulations with the metadynamics algorithm, we had to choose the biasing collective variables. In the second part, we explore the properties of collective variables from a theoretical perspective, setting up a framework to evaluate how suitable any specific variable is to bias efficiently a metadynamics simulation. In particular, we investigate whether their time evolution can be described as the outcome of an effective, overdamped Langevin equation and we develop an algorithm to recover the drift and diffusion coefficients from short time series. We apply this investigation to two variables commonly used to describe the conformational properties of proteins, the dRMSD and the fraction of native contacts. We show that the dynamics in the projected space depends on the microscopic conformations, although the equilibrium properties are compatible with a monodimensional description of the system.