Drug discovery projects involve the identification of hits or lead compounds having favorable drug-like properties. In other words, it is a multivariate optimization problem and the properties to be optimized are binding affinity, binding specificity, ADMET properties, solubility, and bioavailability. Computationally these properties can be estimated using physics-based and data-driven models. In the physics based models, the drug-like properties are computed as free energy differences. In data-driven models, the molecules are described numerically using various descriptors and finger-prints and machine learning and deep learning models can be built using dataset of descriptors/finger-prints and corresponding drug-like properties available from experiments. The physics-based models are usually computationally demanding and so are limited to smaller chemical space. However ML/DL models, can be applied to compute the drug-like properties of huge chemical libraries such as GDB17, Real Enamine DB, ZINC having more than billion compounds. The implementation, merits and demerits of these two approaches will be discussed in some detail in this paper.