Drug discoveries often need expertise knowledge and insanely complex biochemical tests for the discovery of molecular chemical properties. In recent years, there has been an increasing trend of using AI and deep learning-based tools which aid the domain expect to speed up the process of drug design. The use of dynamically un-supervised deep learning systems is used to identify certain properties of atoms of molecules that could have aided the pharmaceutical scientist at many times. The drug discovery comes under sixth goal of millennium development goals which is used as a standard procedure to combat diseases such as HIV, AIDS, dengue, malaria, and another global pandemic. In this paper, we propose a new un-supervised molecular embedding procedure, which provides a continuous vector of molecule in their latent space. These molecules in their latent space can aide in generation of new atoms or a combination of atoms which have relevant chemical nature and can be used as a direct and effective replacement for the existing molecules. The model proposed in this paper is an LSTM autoencoder (long short-term memory) to model the sequence to sequence approach since the molecule input is taken a string format called simplified molecular-input line-entry system (SMILES).