Advanced statistical methods are rapidly impregnating many scientific fields, offering new perspectives on long-standing problems. In materials science, data-driven methods are already bearing fruit in various disciplines, such as hard condensed matter or inorganic chemistry, while much less has happened in soft matter. I will describe how we use data-driven methods to better understand structure-property relationships and move toward material/compound design in soft matter. The first example will consist of the design of polymer membranes with improved gas-separation properties. The training of kernel-based ML model on a relatively small experimental dataset led to the identification and experimental verification of exceptional CO2/CH4 separation performance. Moving to multiscale computer simulations, we explore the use of coarse-grained models in the context of compound screening. Modeling the passive permeation of drugs across a phospholipid membrane, we generate redictions for more than 1 million compounds, and connect key functional groups to the thermodynamic process. Finally, I will describe how we further connect back to an atomistic resolution using deep generative adversarial networks.