The enterprise of modern science evolves alongside its underlying structure, encompassing interactions among scientists, institutions, and culture, as well as the development of scientific ideas and discoveries. The increasing availability of digitized scholarly data, coupled with enhanced computational power, offers unprecedented opportunities to characterize, critically examine, and untangle how these complex, intertwined entities shape the inner workings of science. Through three empirical studies, this dissertation investigates how underlying social and infrastructural elements can influence the production, diffusion, and consumption of scientific ideas and discoveries. The introduction outlines the three studies. The first chapter tackles the question of how tacitly encoded configurations and undocumented components can influence the estimates from randomized clinical trials, demonstrating the role of socio-epistemic bubbles in the production of scientific medical knowledge. The results suggest why meta-analyses may not mechanically resolve scientific disagreements and disputes, contrary to widespread expectations. The second chapter extends the analogy of bubbles to the realm of attention in biomedical scientific knowledge, reflecting the phenomenon of bubbles and collapses in financial asset markets. It reveals that restricted diffusion within social and scientific 'bubbles' can precede sudden collapses in scientific attention, offering a straightforward framework for identifying early signs of these bubbles in science. The third study explores how the presence of code repositories alongside machine learning research affects the citation rates of papers. It finds that the popularity of ML frameworks, such as PyTorch and TensorFlow, used in these repositories can have second-order network effects, underscoring the latent role of technological artifacts and infrastructure in scientific dissemination. The final chapter concludes the dissertation with reflections and outlines future research directions.