Drug discovery is an expensive process. It has been estimated that a new drug compound that is introduced in the market after FDA approval carries a cost of approximately $800 million from the conception of target implicated for a disease to successful identification of chemical entity or drug that is successful in human trials. And this number does not even include expenses incurred for compounds that fail in the later stages of drug discovery due to problems such an toxicity, lack of efficacy in humans (although it shows potency in in-vitro experiments), poor physical properties that make it unsuitable for absorption, distribution at the required site, metabolism or excretion in human body. There is an urgent need to cut the cost of developing new drugs (to bring overall cost lower for the producers and consumers alike) by identifying promising candidate drug compounds in the early stages of drug discovery. In order to achieve this objective, in recent years the development of computational techniques that build models to correctly assign chemical compounds to various classes, to retrieve potential drug-like compounds (hits), or to identify promising new targets for different diseases has been an active area of research.

In this thesis we study and develop algorithms for various problems in the field of computational drug discovery that show state-of-the-art performance. Specifically, we develop methods in three key areas of drug discovery - representation of chemical compounds for classification and retrieval (termed as Effective Descriptor-Space Representation Problem), identification of diverse set bioactives for a given query (termed as Scaffold-Hopping Problem) and identification of likely targets for given (set) of chemical compounds (termed as Target Identification Problem). All of these methods utilize experimental data on chemical compounds binding to various assays and is derived from in-vitro as well as in-vivo experiments. The methods we propose are inspired by research in the areas of information retrieval and machine learning. Our extensive experimental evaluation shows that most of the methods developed in this work are either competitive or substantially outperform previously developed approaches to solve the above problems in drug discovery.