The understanding of molecular effects in nanoscale environments is becoming increasingly relevant for various emerging fields. These include spectroscopy for molecular identification as well as in finding molecules for energy harvesting. Theoretical quantum chemistry has been increasingly useful to address these phenomena to yield an understanding of these effects. In the first part of this dissertation, we study the chemical effect of surface-enhanced Raman scattering (SERS). We use quantum chemistry simulations to study the metal-molecule interactions present in these systems. We find that the excitations that provide a chemical enhancement contain a mixed contribution from the metal and the molecule. Moreover, using atomistic studies we propose an additional source of enhancement, where a transition metal dopant surface could provide an additional enhancement. We also develop methods to study the electrostatic effects of molecules in metallic environments. We study the importance of image-charge effects, as well as field-bias to molecules interacting with perfect conductors. The atomistic modeling and the electrostatic approximation enable us to study the effects of the metal interacting with the molecule in a complementary fashion, which provides a better understanding of the complex effects present in SERS. In the second part of this dissertation, we present the Harvard Clean Energy project, a high-throughput approach for a large-scale computational screening and design of organic photovoltaic materials. We create molecular libraries to search for candidates structures and use quantum chemistry, machine learning and cheminformatics methods to characterize these systems and find structure-property relations. The scale of this study requires an equally large computational resource. We rely on distributed volunteer computing to obtain these properties. In the third part of this dissertation we present our work related to the acceleration of electronic structure methods using graphics processing units. This hardware represents a change of paradigm with respect to the typical CPU device architectures. We accelerate the resolution-of-the-identity Moller-Plesset second-order perturbation theory algorithm using graphics cards. We also provide detailed tools to address memory and single-precision issues that these cards often present.