Monday, November 26, 2012

When n=1: How can we encourage scientific reproducibility?

High profile reports of bacteria that incorporated arsenic instead of phosphorus and of a particle traveling faster than the speed of light have not fully stood up to public scrutiny. The inability of other scientists to replicate these widely publicized experiments has brought increased attention to the issue of the reproducibility of scientific experiments.
Adding fuel to the fire, Amgen scientists reported they could not repeat 89% of the published findings on promising targets for cancer therapeutics they investigated. These events have led to outrage that public dollars are being spent on such poor research. There have been a number of proposals for ways the scientific community can maximize the reproducibility of published results. These include additional ethics training for students and young investigatiors and standardization of guidelines for publication. Unfortunately, too often the lack of reproducibility gets conflated with the more serious issues of carelessness and fraud. While both poor science and (in rare cases) outright fraud contribute to the publication of work that cannot be reproduced, there are other issues to consider.

It is important to be clear that the failure of one lab to replicate the work of another is not the same as proving the original work to be false. Many bench scientists struggle to reproduce results, both published and within their own labs. When dialogue is open between the two scientists performing the experiments, it is usually easy to see where miscommunication or lack of detail in a protocol has led to a different result. In my opinion, a vital step in reducing issues with reproducibility is to encourage the publication of detailed protocols. Far too often, Materials and Methods sections are short and among the first areas to be cut when conforming to a journal’s word limit. Instead, we should expect each published article to clarify important details including the temperature at which experiments were performed, the concentration of all reactants and the equipment used for each step of a procedure. Only when replicate experiments have been performed precisely under the same conditions should the original be regarded with skepticism.
New and interesting ways of providing detailed experimental procedures have proliferated in recent years with the publication of Nature Protocols and JoVE, two repositories for highly detailed methods. Providing thorough explanation of techniques and procedures will become common practice if high profile labs lead the way by sharing their novel methods. The NIH can encourage the use of these repositories by making procedural transparency a component of the score that determines whether a grant is funded or not.
There are creative attempts going on to identify results that are irreproducible (or conversely, identify reproducible results) to minimize time and effort spent in other labs following up on poor data. The blog Retraction Watch wants to make sure the scientific community is aware of papers that have been withdrawn or retracted. While this project less directly aids in improving reproducibility, it helps with the larger goal of preventing the waste of time spent by researchers trying to replicate false or incomplete experiments. The authors of the blog note in their FAQ section that there is no comprehensive database of retractions from scientific journals. While the retraction of an article may be noted in the place of the original manuscript on the publisher’s website, little publicity is given to these notices. Indeed, it is hardly in the publisher’s best interest to do so. 
A particularly bold group has approached this problem by founding the Reproducibility Initiative, a new resource to help scientists by adding to the impact of experiments that have been reproduced. For a fee, the group will match interested investigators to researchers who will then attempt to repeat their experiments. As part of the service, the investigator has the option to publish the results of validating experiments in an online journal (the publication is optional so investigators may choose not to publish experiments that conflict with their original results). Validation of the initial results qualifies the original manuscript--if published in participating journals--for recognition of that reproduction. Presumably, this validation adds to the impacts of the results.
How well this initiative succeeds will depend entirely on the quality of the scientists performing the follow-up experiments and the ability to communicate between the original and replicating labs. If the follow-up lab is able to quickly replicate a result, the situation will be beneficial to all. However, if the follow-up lab cannot replicate the published result, the amount of benefit will depend on the two labs working together to determine why the replication failed. An inexplicable discrepancy helps no one, for the reasons discussed above. As a scientist, I would certainly be glad to know others could reproduce my results, but if they failed to do so, I would not necessarily trust their results over my own due to my greater knowledge of my own methods. Inexplicable discrepancies could lead to potentially time consuming and costly searches to reconcile the results of the two labs. This may prevent scientists from using the service if they have even one bad experience.
How do you think the scientific community can improve the reproducibility of publically funded research? Leave your ideas in the comments!

Irene Reynolds Tebbs 
6th year, Molecular Biophysics and Biochemistry