I just came back from ICSB 2013, the leading international conference on systems biology (short write-up here). During the conference Bernhard Palsson gave a great talk, which he ended by promoting a view that (I suspect) is widely held among computational and theoretical biologists but rarely vocalized: most high-impact journals require that novel predictions are experimentally validated before they are deemed worthy for publication, by which point they cease to be novel predictions. Why not allow scientists to publish predictions by themselves?
This is an issue that frustrates many non-experimental biologists, myself included. There is an unspoken (and sometimes not quite unspoken) class distinction in biology that separates theoretical and experimental work, with the former seen as inferior (not “real” biology). It is akin to the divide between theoretical and experimental physics, except the situation is reversed in biology.
My first reaction was to concur with him, but on further reflection I think that the problem is deeper. To be sure, there are dubious sociological factors that have partly driven and continue to drive this separation, for example the general math phobia of many experimental biologists. But it would be disingenuous to pretend that the problem stops there. The real issue, in my opinion, is that making predictions in biology is cheap, and it is so in multiple ways.
Sociologically, most biological predictions are so laughably bad that no one feels embarrassed by having made a wrong prediction, and so there is no social mechanism that makes people stop and think before making predictions, lest their reputation suffers. As a result most predictions are (justifiably) not taken very seriously, and in some sense the good is lumped with the bad. Compounding this problem is a lack of biological depth and understanding by many non-experimental biologists (a situation that is improving, just as the mathematical sophistication of experimental biologists is increasing). As a result much of theoretical biology occurs in a vacuum that is devoid of an understanding of the basic biological phenomena. Contrast this with physics, where theoretical physicists have an equal if not deeper grasp of the phenomena than their experimental counterparts.
Technologically, most predictions are literally cheap, in that they consume little computational power and so require minimal financial resources. In cases where this is not the case, for example very long time-scale molecular dynamics simulations, the predictions do get more attention, in part I think because people unconsciously want to give credit “for the effort”, and technological tour de forces are often fetishized. (An example of a similar phenomenon is the mathematical machismo of economics, on which much has been written.)
Finally and most perniciously, predictions are scientifically cheap. This is the most interesting one and deserves further attention. The underlying problem is the lack of theory or phenomenology in biology. Predictions can and are made in a vacuum. There is no overarching theoretical structure that constrains predictions and guarantees their internal consistency. Another way of saying this is that there are only models in biology, and more importantly they are not phenomenological models, because there is no theory that they must adhere to. They are literally things that people just make up. Phenomenological models on the other hand must pass the test of being consistent with theory, and theory must pass the test of being correct and predictive over a broad range of phenomena, internally consistent, non-trivially generalizable, and aesthetically minimal.
The contrast with physics is instructive, where the situation is markedly different. One of the most stunning examples of theoretical work is Paul Dirac’s prediction of the positron. If Dirac had simply postulated the positively charged twin of the electron out of thin air, no one would have taken him seriously. Instead, the prediction was made within the context of an extension (relativistic quantum mechanics) of a very promising theoretical framework (quantum mechanics). This theoretical framework explained very many things, and did so with exceeding quantitative accuracy. Equally importantly, it was an internally consistent mathematical structure that one could not simply hack things onto in an ad hoc fashion. Dirac’s extension was hard, in the sense that it required significant technical effort for it to work, and the prediction that as a result of this principled extension of quantum mechanics a new particle had to exist carried weight, precisely because of the principled nature of the extension. In some ways the situation that exists today in physics in which theory is elevated over experiment was a result of such breathtaking successes of 20th century theoretical physics. An ironic byproduct of this can be seen in the recent faster-than-light neutrino debacle. So strong were the theoretical objections to the experiment that most physicists did not believe it to be correct, and as expected it eventually turned out that experimental error was to blame. Theory trumped experiment.
Biology has no real parallels, except perhaps in the qualitative theory of evolution. It is appropriate then that the one recent glaring exception to my claim is this Cell paper, very much a high-profile prediction, and one based on utilizing evolutionary principles. Beyond a handful of examples however most biological predictions are scientifically shallow, escaping the requirement of having to satisfy a rigid mathematical structure and thus difficult to judge based on inherent theoretical merit. The basic checks that exist in more rigorous fields, sociological, technological, and theoretical, are lacking in biological predictions and this devalues their currency. People don’t trust cheap things.
What is the solution? I suspect that the first two issues will ultimately be fixed. As predictions start getting better, people will start taking them more seriously, which will increase the onus on predictors to make accurate predictions. Similarly, computation is beginning to overtake the experimental cost in some fields, such as DNA sequencing. When this becomes more widespread, biologists will begin to see computational predictions as serious investments. It is the third issue however that I believe will remain a challenge for some time to come, perhaps forever. Developing a traditional theory of biology, as I fantasized long ago, is probably impossible. But I think there will be something else in its place, a topic to which I will return in the future.