Full Computational Reproducibility in Biological Science: Methods, Software and a Case Study in Protein Biology

Independent computational reproducibility of scientific results is rapidly becoming of pivotal importance in scientific progress as computation itself plays a more and more central role in so many branches of science. Historically, reproducibility has followed the familiar Popperian [38] model whereby theory cannot be verified by scientific testing, it can only be falsified. Ultimately, this implies that if an experiment cannot be reproduced independently to some satisfactory level of precision, its value is essentially unquantifiable; put brutally, it is impossible to determine its scientific value. The burgeoning presence of software in most scientific work adds a new and particularly opaque layer of complexity [29]. In spite of much recent interest in many scientific areas, emphasis remains more on procedures, strictures and discussion [12, 14, 16, 29, 30, 37, 41], reflecting the inexperience of most scientific journals when it comes to software, rather than the details of how computational reproducibility is actually achieved, for which there appear to be relatively few guiding examples [6, 10, 17]. After considering basic principles, here we show how full computational reproducibility can be achieved in practice at every stage using a case study of a multi-gigabyte protein study on the open SwissProt protein database, from data download all the way to individual figure by figure reproduction as an exemplar for general scientific computation.