Abstract
Background Computational biology provides widely used and powerful software tools for testing and making inferences about biological data. In the face of rapidly increasing volumes of data, heuristic methods that trade software speed for accuracy may be employed. We are have studied these trade-offs using the results of a large number of independent software benchmarks, and evaluated whether external factors are indicative of accurate software.
Method We have extracted accuracy and speed ranks from independent benchmarks of different bioinformatic software tools, and evaluated whether the speed, author reputation, journal impact, recency and developer efforts are indicative of accuracy.
Results We found that software speed, author reputation, journal impact, number of citations and age are all unreliable predictors of software accuracy. This is unfortunate because citations, author and journal reputation are frequently cited reasons for selecting software tools. However, GitHub-derived records and high version numbers show that the accurate bioinformatic software tools are generally the product of many improvements over time, often from multiple developers.
Discussion We also find that the field of bioinformatics has a large excess of slow and inaccurate software tools, and this is consistent across many sub-disciplines. Meanwhile, there are few tools that are middle-of-road in terms of accuracy and speed trade-offs. We hypothesise that a form of publication-bias influences the publication and development of bioinformatic software. In other words, software that is intermediate in terms of both speed and accuracy may be difficult to publish - possibly due to author, editor and reviewer practices. This leaves an unfortunate hole in the literature as the ideal tools may fall into this gap. For example, high accuracy tools are not always useful if years of CPU time are required, while high speed is not useful if the results are also inaccurate.
Competing Interest Statement
The authors have declared no competing interest.
Footnotes
We have added some additional GitHub features to the analysis, clarified a few issues and expanded the Conclusions section. We have reduced the number of extraneous figures from the Supplement too.
https://github.com/Gardner-BinfLab/speed-vs-accuracy-meta-analysis
