The importance of polypharmacology is beginning to be recognized in the field of drug discovery. The effectiveness of the existing drugs are increasingly being attributed to inhibition of multiple targets in a number of cases. However, there development of approaches and methods to rationalize the process of polypharmacological target selection is yet to receive sufficient attention. Here we propose a structural proteomics approach that utilizes the structural information of the binding sites at a genome scale to first explain the basis of polypharmacology of known candidates and then predict new polypharmacological targets in Mycobacterium tuberculosis (Mtb). Computational methods have been used for proteome-wide binding site detection, comparison and functional characterization to construct the ‘pocketome’ of Mtb. For about 13858 binding sites in the proteome, all pair similarities have been computed, which are then used to obtain clusters of targets, each cluster containing a set of similar binding sites and theoretically having a potential of being modulated by a single drug. The clusters are then matched onto existing drug-binding sites from DrugBank to obtain sets of possible lead compounds. A therapeutic polypharmacology index (TPPI) has also been designed to estimate the therapeutic polypharmacology potential at the level of binding sites. A binding-site similarity network has been used to construct high-confidence drug-target network consisting of important target proteins derived from systems analysis and binding sites of approved drugs. This study identifies putative targets with polypharmacological potential along with possible drugs for repurposing and use as lead compounds for inhibiting Mtb. The method is fairly generic and can be adapted in other drug discovery programmes.