Drug discovery in the 21st century is expected to be different in at least two distinct ways: development of individualized medicine utilizing genomic information and emergence of an integrated in silico protocol for facilitating target identification, structure prediction and lead discovery. The expectations from computational methods for developing suggestions on potential leads reliably and expeditiously, are continuously on the increase. Several conceptual and methodological concerns remain before an automation of lead design in silico could be contemplated. The novelty of the candidates generated, their geometries, the partial atomic charges and other force field parameters for enabling energy evaluations is one concern. A proper account of the flexibility of the candidate molecule and the target, a consideration of solvent and salt effects in binding and a reliable methodology for developing quantitative estimates of binding affinities is another. Finally the drug-likeness of the candidates generated is yet another concern. Each of these issues warrants a careful consideration. In this review, we sketch a system independent, binding free energy based, comprehensive computational pathway from chemical templates to lead-like molecules, given the three dimensional structure of the target protein and a definition of its active site, focusing on some emerging in silico trends and techniques. We survey current methods for generation of candidate molecules and some popular protocols for docking candidates in the protein active site. We discuss the theory of protein-ligand binding in the rigorous framework of statistical mechanics and assess the current strategies for affinity based filtering of candidates. We address concerns related to flexibility of the target and the candidate, solvent and salt effects in lead design. We present a realization of the pathway proposed in a high performance computing environment for cyclooxygenase-2 target wherein the computational protocols could sort drugs from nondrugs, assuring the viability of the overall strategy. We highlight a few case studies indicating the current level of agreement between theory and experiment in eliciting binding affinities. Finally, we present a critical assessment of the computational steps involved in binding affinity based active site directed lead molecule design and further improvements envisioned for potential automation.