A recent issue of Science magazine touted the utility of computer models for understanding complex biological systems. They praised computer simulations of the cell cycle and signal transduction for the "unique insights" they provide to cell biologists. The Hughes Medical Medical Institute even announced the availability of investigator positions in "computational biology." So I suppose that it is now official: it is acceptable for biologists to use computers.

Not that we haven't used them for years. We write our papers on them, draw a few figures, design our posters, and look up references on the net. A few biologists even use spreadsheets and databases to crunch some numbers, do a few calculations, and keep track of where their cell lines and reagents are located. But as a whole, biologists have never really liked computers. Sure, they can be useful, but if you have to use a computer in your work, well then, you really aren't a biologist.

At least this is the line that I have been getting for years. As someone who, well, actually likes computers, I have frequently had trouble getting my peers in molecular and cell biology to overlook my computer habit. I found it necessary to convince reviewers of my papers and grants that I earn my results the old-fashioned way - I run gels.

I suppose my amusement at the attention that computers have received recently in the trendy scientific journals is partially due to my experience in having to hide my computer fascination for the last 20 years or so. I certainly understand why computers have recently become so hot. It is a combination of concern over the direction of "post-genomics" science as well as the great impact the Internet has made in the way biologists now ply their trade. Certainly, the explosion of genetic information has transformed the computer from a useful tool to an essential partner in research. Even contemplating going back to the old ways of looking up journal references and scanning gene sequences is enough to send shudders through the most computer-phobic biologist. Now a new idea is starting to take hold. If computers are so great at doing references, then perhaps they can help us understand data.

I suspect that although the new enthusiasm for computers in biology is genuine, it overlooks some basic problems in implementation. The basic difficulty, as I see it, is that although biologists use computers, they do not trust everything that comes out of them. It is one thing to use them to print up nice-looking graphs, but it is an entirely different matter to use them to think better. Biologists also seem reluctant to accept results obtained from computer models. This likely stems from basic human nature and the inherent skeptical nature of scientists. If I devise a computer model that claims that the sky is blue, no one will object. Of course, I could not publish the model because it shows a self-evident truth. If my model shows that the sky is actually black, but we only perceive it to be blue, then I may have a problem publishing that model as well. Reviewers will only accept models of nonintuitive results if they can understand how the conclusions were reached. Otherwise, how can they distinguish between genius and misinformed enthusiasm? This is the crux of the computer problem. Unless biologists understand how computer models work, they will not accept them. If they are good scientists, they have no other choice.

Of course, not all computer models are good. There were a number of poorly conceived models published many years ago that generated more controversy than insight. (Does anyone remember the "probabilistic transition" model of the cell cycle?) However, these early failures were mostly due to a lack of mechanistic understanding of the systems, and not due to an inherent fault of modeling. Certainly, the field of molecular and cellular biology has its own share of spectacular faux pas, but I don't see anyone blaming techniques (just think "two-hybrid screen"). Somehow, the distinction between the medium and the message does not hold for computer models.

My first experience with this issue goes back to an incident 18 years ago. I had recently purchased an Apple II computer and had learned how to program in BASIC. Back then, there were few software packages available for personal computers. It was understood that you got a computer so that you could program the darned thing. This was part of the experience. A computer was a new toy that allowed you to burn all your free time in trying to program it. My madness had a method, however. I bought my computer so that I could model a biological process. It is important to understand that even 18 years ago, computer modeling had been around a long time. The problem was that if you wanted to do something as esoteric as use a computer for a biological problem, you had to hunt down a mainframe and convince its caretaker to let you use it. Pretty stiff entry requirement. The great thing about the Apple II was that a normal laboratory could actually afford one. Of course, you had to convince the P.I. that the personal computer was actually useful in research. I, obviously, was more talented in programming than in persuasion. Hence I ended up paying for the computer myself.

It was a wonderful investment. Within about six months of buying the computer, I had put together a program that allowed me to simulate the behavior of the EGF receptor. To write the program, I had to put some equations together that described receptor dynamics. I based my models on some equations from my biochemistry textbook and made good use of a computer calculus course I had taken as an undergraduate. The remarkable thing about the whole model was that it worked. My computer simulated the behavior of receptors quite well, and the equations appeared to be genuinely useful. The whole endeavor had a whizbang quality that I loved. Better yet, my paper describing the computer project got published in the journal Cell. Benjamin Lewin always had a soft spot for weird science.

To say that I was psyched on computers at the time was an understatement. I thought, "I am going to change the way biology is done." For the first time, "normal" biologists could use computers to probe their data for novel insights. What I did not understand at the time, however, is that models generated by novel approaches are only acceptable when they support current views.

The immediate goal of the laboratory I was working in was to isolate the thrombin receptor. Standard approaches to receptor purification did not work on this system. This prompted a bright graduate student in the lab to suggest that perhaps thrombin just clipped a specific protein at the cell surface and what we were seeing as "receptors" were simply irrelevant binding sites. This seemed a pretty radical suggestion at the time. We, therefore, decided to settle the matter with some critical experiments in conjunction with a mechanistic computer model that made specific predictions for "bind and clip" versus "clip only" hypotheses. After many long months of experiments, we found that our data matched the clip-only model exactly. This suggested that current dogma might be wrong.

We felt our results might interest other scientists in the field. Surprisingly, we found we could not publish our study. It was sent to journal after journal and kept coming back with the same opinion that "far more justification than a computer model was needed to overthrow current dogma". Quantitative biology was seen as an oxymoron.

This is where we learned an important rule about scientific revolution: if you want to dismantle a current paradigm, you have to have a new one to take its place. Although our computer model could predict the characteristics of a genuine thrombin receptor, it could not actually isolate the molecule itself. Without such a magical capability, our model was felt to be of limited usefulness.

Our lab advisor consoled us following our rude rejection by the scientific establishment. He pointed out that by failing to consider our findings, they were just hurting themselves. We would not be so stupid, and so we would use the models to plan our own experiments. If models were useful, then we would benefit from using them ourselves.

Our data was eventually published sans the computer model. Years later, a clever investigator finally identified the thrombin receptor by using expression cloning in frog oocytes. The receptor had exactly the properties predicted by our computer model. This was personal vindication of a sort. My advisor's lab did not waste any additional effort on purifying the thrombin receptor, but instead, put its efforts to good use in studying less controversial subjects such as EGF receptors and protease inhibitors.

I have since teamed up with scientific collaborators who share my enthusiasm for computer models. We occasionally write papers that use models and still must fight to get them published. Despite their poor reception by my peers, I still like computer models. Experience over the years has convinced me that they can be extremely useful tools. Unfortunately, I am also convinced that most mainstream biologists are still not ready for them. To trust a tool, you have to be aware of its limitations and advantages. But I see few biologists being trained to use computers to analyze biological data. There are probably fewer biologists learning to use computers as analytical tools today than 20 years ago. Back then, they were new and exciting. Today, they are pieces of furniture.

Francis Crick was once quoted as saying that no biologist had ever made a discovery using a mathematical model. I would reply that no biologist has ever made a discovery by running an electrophoretic gel. They make discoveries by using their brains. Computers, like all scientific tools, are only as good as the person who uses them. If biologists don't understand how computer models are constructed, they won't know their strengths and limitations. Without some foundation of trust, biologists will be unlikely to utilize or accept this powerful method of data analysis.

It has been predicted that computers can help us understand the vast amount of biological information that is being generated at an ever increasing rate. If this promise is to be realized, biologists must become more familiar with computers. This concern is more than a simple desire for mainstream acceptance. Without the cooperation of biologists, modelers will have few options for testing predictions. Extracting parameters from the literature is a poor alternative to having a collaborator generate data on command. But what is the best way to familiarize biologists with modeling? I am really not certain. Most negative attitudes are probably due to lack of exposure. The number of serious modeling efforts being published, however, are definitely on the upswing. In fields such as structural biology and genomics, computers are already essential tools. Perhaps as biologists see more examples of how computers can enhance the imagination rather than replace critical thinking, trust in them will build. Then computational biology will cease to be just a buzzword and will truly join the mainstream of biology.