SymbolsAndArrows

The units of a first order rate constant

2011-11-02T10:13:00.000-07:00

One of our principal jobs as kineticists is communication. Only a minority of our experimental colleagues is completely comfortable with the physical and mathematical concepts that underpin a kinetic model of a biological system, and just as we rely on the experimentalists to teach us key biological principles, they rely on us to teach physical concepts.

Every kineticist presenting a biological model has encountered a member of the audience who does not comprehend the meaning or significance of a first order rate constant. Often the communication hurdle boils down to confusion about units. People have trouble with inverse time units like sec^-1 or min^-1.

One approach to this communication problem is to speak in terms of mean residence times. This is fine if your compartment/species/state has only one exit pathway, but it will just confuse people even more if there are two or more. Another approach is to define a rate constant as the fraction of molecules in the source compartment that leaves by this pathway per unit time. A few more people will understand this definition.

But my favorite solution to this dilemma is to WRITE a rate constant in the usual form, for example 0.035 min^-1, but when reading out loud or speaking, SAY "3.5 percent per minute."

People are fine with units like m/sec or L/min where the units have both numerator and denominator. It's when there is ONLY a denominator unit that they have trouble.

So do your audiences a favor and quote your first order rate constants in percent per min or percent per second units. There is no real difference, of course, between 0.035 min^-1 and 3.5 percent per min, but communication, you will find, is increased 100%.

Occasionally, someone in the audience who has some understanding of kinetics will insist that your units are incorrect. This has happened once in the 15 years I've been taking this tack.

Try it. I think you'll find that the increase in understanding and appreciation of your work is well worth this small risk.

Merging expertise

2010-09-04T10:50:00.000-07:00

Recent decades have seen a wave of mergers in the pharmaceutical and biotech industries, and this consolidation is forecast to continue into the 2010's.

There are probably hundreds of reasons why companies decide to merge with or to purchase other companies, but for R&D organizations in general and for pharmaceutical firms in particular, mergers are occasions for integrating expertise. What, we might ask, is the most efficient and profitable way to merge expertise?

To me, the task of merging expertise ought to be a modeling process. Two groups of scientists are brought together by a merger. No matter how it was decided who should be part of each new R&D team, everyone sees an infusion of new blood - new ideas, new concepts, new beliefs. This newness can precipitate conflict or insight. In my experience, modeling is the best tool for assuring that the result of the merger is insight, not conflict.

Pharmaceutical R&D teams always need a big-picture view of the disease they aim to ameliorate. But again and again, we find that even within a single research group - especially one with several new members - individual scientists hold very different big-picture views. These differences have practical importance because, often, the best lead compound or the best next experiment depends on which of those big pictures you believe.

Unfortunately, the choice of big-picture view is often driven by who is the PI, or what the boss says, or who is most articulate, or whose native language is the language of the group. Might it be more productive if, instead, this decision was based on scientific hypothesis testing?

Modern modeling tools can offer objective, low cost, purely computational evaluation of all those disparate big-picture views. For example, you can, as a group, agree on the full set of experimental results that your group's big-picture model should explain. Then you can run the corresponding experiments on each individual's big-picture model. Models that fail this test can be abandoned with good reason, and models that succeed now have strong empirical support, not just someone's considered opinion of their worth. Everyone can agree that models passing such tests are worthy of being moved forward.

If two successful models have different underlying biological mechanisms, then modeling can be used to design an experiment that differentiates them.

At the end of this process, which takes about 6 weeks, you have built real intellectual property. You have an empirically tested framework for designing experiments, evaluating results, choosing targets, and selecting lead compounds. You have, in effect, a strong foundation for mechanistic, evidence-based pharmaceutical development.

This approach can have many benefits. It reduces friction among team members. It provides clear evidence upon which a scientific team can make confident, empirically supported recommendations to its managers. And, to the point of this post, modeling effectively leverages the expertise and accumulated biopharmaceutical wisdom of ALL the members of a merger-produced team.

Tackling Complex Diseases

2010-06-23T10:48:00.000-07:00

We are surrounded by evidence that major diseases are not yielding to the hard work and sustained efforts of basic, clinical and pharmaceutical scientists. Despite major progress, heart disease, vascular disease, cancer, diabetes, and dementia are still major threats to quality of human life.

I think these diseases have one thing in common. They are far more complex than the diseases for which humankind has cures. The word complexity is on everyone's lips, but biomedicine is struggling with what to do about it.

This is actually a great paradox. The physical sciences and the engineering sciences have known for three centuries that the only effective tool for analyzing and controlling complexity is mathematics. In the last century the word mathematics has been augmented by computation and become even more powerful. But still, neither mathematics nor computation is considered part of the mainstream of biomedical research.

This is not to deny the rise of biomedical engineering and systems biology, but merely to point out that these disciplines are revered for their technologies, not for their cures or even for their pivotal insights into complex disease. I think this is because biology and disease are, fundamentally, not engineering problems. They are mysteries. We need Sherlock Holmes, not Captain Nemo.

Biologists implicitly, and sometimes explicitly, contend that complex theories are not testable and so they persist in testing simple theories. But simple theories, elegant as they are, have no chance to solve complex diseases. I say that modeling makes complex theories testable; this is why modeling is the future of biomedical research.

Marketing multidisciplinarity

2010-04-03T09:57:00.000-07:00

A recent insightful post on the LinkedIn Computational biology Discussion board came from a post-doc at Rutgers, Michael Andrec. His query generated many thoughtful replies. Here is his post, followed by my own thoughts on the subject

How can truly interdisciplinary scientists describe and market themselves?
This is a question that's looming large for me at the moment. While my degrees are from chemistry departments, my post-PhD research has been so diverse that I no longer fit into any neat pigeon-hole. So, while I have substantial skills and experience in areas that might be labeled "bioinformatics", "structural biology", "statistics", or "computer science", I don't have a degree or an extensive enough track record in any one of them to feel comfortable calling myself a "bioinformatician", "structural biologist", "statistician", or "computer scientist". I'm none of them and all of them, to put it in a Zen-like way.

This makes it harder to find a new position (since ads are usually framed in terms of one the standard pigeon-holes), and it also makes it a challenge to describe oneself. If I had a CS degree, for example, I might be able to say "I'm a computer scientist with interests and skills in biology", and most of you would instantly know where I was coming from. I find it very hard to describe my professional self in a concise way like that...

Am I the only one with this conundrum? I'm sure there must be more of us out there...

Here is my response. If you have additional or contrary ideas, by all means add them to the LinkedIn discussion, which can be reached by clicking on the "Marketing multidisciplinarity" title of my blog post.

This is an exceptionally perceptive thread. To put my remarks in context, I'm an MIT electrical engineer with a PhD in Physiology who was for many years a professor in physiology and biomedical engineering at Hopkins before leaving to co-found a systems biology software/consulting startup.

My view is that traditional disciplines are the anachronistic legacy of reductionist science, but that they will remain centers of power for another generation or two simply because so many people and institutions and societies and journals have vested interests. This is not a criticism; it's just a fact.

On the other side of the coin, I think it's fair to say that smart people everywhere are aware that we have left reductionism behind and are already well-embarked on synthesis. There are new departments and new societies and new journals and new companies and new LinkedIn groups that are defining this new reality. But pioneers have always faced the problems itemized in this thread. You cannot be a pioneer and remain in the comfortably established career tracks. Being a pioneer involves risk.

But the reward is that you can build a career that YOU think is important - a professional life that YOU think is worth living. In truth we are enormously privileged to have this opportunity.

The advice one gives depends on career goals. If you want to be an academic, look at those new departments with poly-hyphenated names. This is how academia hedges its bets. Molecular Integrative Physiology, Systems Biology, Biochemical Engineering, Chemical Systems Biology, or anything with "computational" or "informatics" on the department nameplate. Someone with a vision has worked hard to create these islands of innovation within an academic organization that is notoriously difficult to change. Academic hopefuls should also consider the journals in which they publish and the journals where current faculty members in your ideal department are publishing. What meetings do they go to? Go to those meetings. Pay your own way if necessary.

If you are not aiming at academics, you have a lot more freedom but you will also need to be convincing more often, because people pay for results not skills.In my view, the physicists who were so successful in the early days of molecular biology are remembered for their spectacular scientific results, not for the fact that they were PhD physicists.

I see too many resumes with a long list of, say, software skills and no indication of what the applicant has actually accomplished or published. Science, whether academic or for-profit, is not about accumulating skills. Science is about solving problems that really matter. Everyone on this thread knows this; you are all learning what you need to learn to solve the problem at hand.

Solve a seriously important problem and you will have no shortage of opportunities. Sources of jobs and/or funding will say to themselves, "Look at what this person has already done. I'm willing to bet they can do more great things."

If I could change one thing in modern science, I would stop the fragmentation and specialization. To do synthetic work, we need more broadly based professional societies, academic departments, and journals, NOT more specialized ones. Current academic department structures evolved from the reductionist paradigm and simply do not support synthesis effectively. One might imagine that the AAAS could serve this synthetic role, but they have become so focused on scientific policy and interactions with government that no one goes to their meetings to hear the latest science.

It's probably true that a synthesis-focused scientific meeting would be too unwieldy for a physical get-together. Someone needs to think hard about how to do an integrative scientific meeting on the web, with great search capability so each of us can FIND the people we need to talk to.

Obviously your thread resonates with me too.

Pioneering requires courage. Just do it.

Patents on DNA sequences

2010-03-31T11:19:00.000-07:00

From my perspective, DNA sequences are not the right intellectual property upon which to build a biotech firm. Ultimately, sequences yield correlations and correlations rarely identify cause and effect. Myriad might be wise to re-invest its enormous expertise in understanding the control of transcription for the genes it has identified as pivotal. Great companies are flexible.

iHOP

2009-04-26T08:25:00.001-07:00

Investigators whose scientific safari takes them into a dense jungle of proteins, genes and signaling pathways owe it to themselves and their funding agencies to check out Robert Hoffmann's iHOP. When I was in college this acronym stood for International House of Pancakes, but the scientific iHOP stands for Information Hyperlinked Over Proteins and is even more satisfying. This free tool is an ingenious combination of database and text mining that I use for every project I work on. It's especially helpful when your work takes a sudden turn into a portion of a signaling network you haven't worked with before.

The most useful feature, in my view, is that text-mining results yield complete sentences extracted directly from the scientific literature and referring to the gene or protein of interest. Each sentence is followed by a link to the abstract, and every list of sentences can be sorted, filtered, or searched. Hoffmann's concept is clean, simple and powerful. New and useful features appear regularly, and Hoffmann is a thoughtful and insightful correspondent in response to e-mailed questions and suggestions.

For targeted searching of the scientific literature, iHOP beats Google. Try iHOP. I bet you bookmark it.

The Audacity of Synthesis

2009-02-15T16:20:00.000-08:00

As a former academic with a passion for quantitative biology, I retain considerable interest in the organizational approaches taken by the world's universities. What surprises me is that in an age almost universally described as an era of synthesis, we continue to build academic departments as if reductionism was still the order of the day.

I remember years ago smiling at the umbrage taken by biochemistry grad students at the annual meeting of the Federation of American Societies for Experimental Biology when they discovered that their badges were emblazoned with the number "2" while those of us from the no-longer-dominant American Physiological Society displayed the proud "1." They were disconcerted to discover that ASBC had started as a splinter group within APS. Indeed, only rarely have groups of academics who develop a novel shared perspective chosen to remain within their parent organization. Young turks would not be turks if they ceded leadership positions to their elders.

This rebelliousness has provided the founding impetus for countless disciplines. I want to argue that the time has come for retrograde rebellion. Today's young turks should be merging disciplines, not creating new ones.

Unfortunately, these young turks are still reading the "Young Turks' Handbook" they found in their chairman's desk drawer. They act as if the only way to be a young turk is to form a splinter group and make it powerful. Biophysics, Mathematical Biology, Biological Engineering, Biomedical Engineering, Bioengineering, Systems Physiology, Systems Biology, Integrative Bioinformatics, Computational Biology, Physical Biology, Theoretical Biological Physics. What really separates these "disciplines?" What are these Turks thinking?

Dick Ross, one of the great deans at The Johns Hopkins School of Medicine, once floated the idea of reducing all our departments to two: basic science and clinical. In retrospect, it's a brilliant idea. At the time, of course, the faculty and the department chairs were aghast. We suspected he was just tired of negotiating with 30 brilliant adversaries about how to best allocate scarce resources. In hindsight, who can blame him. But you won't be surprised that the established academic fiefdoms rose up en masse to resist the inevitable loss of power. Today, however, a more compelling case can be made for Dr. Ross' idea.

Indeed, the rebels of the next few decades might be well-advised to work toward unification, not further secession. Where are the leaders bold enough and confident enough to work together instead of working apart? Where is the biomathematics department that is willing to join a cell biology department, or the biochemistry department willing to be merged with a physiology department? In the past, such events have always been seen as deaths. We need enlightened leaders who see them as births and who run the merged department with the best interests of ALL its members at heart.

The same could be argued for scientific societies, scientific journals and scientific meetings.

If the 21st century is to become the age of biological synthesis, then our academic structures might usefully reflect this. Academic departments proliferated during the age of reductionism. Perhaps they should contract during an age of synthesis.

Leaders in biological synthesis will recognize they need the perspectives of ALL the former reductionists in order to succeed. What better way to achieve this than to bring all those reductionists together again, in their academic departments, their professional societies and their national meetings?

Courage

2009-02-11T11:21:00.000-08:00

Much of experimental biomedical research is published without any reference to modeling. And we all know modelers who publish without any reference to experimental data. More and more, however, investigators on both sides of the fence are calling for interaction.

The two cultures seem to resist this interaction, and we have all heard, or even spoken, the various rationales for remaining aloof, but I'm wondering if the real problem is a dearth of courage.

Modern science has placed a huge premium on "being right." Investigators often go to extreme lengths to defend their conclusions or their theories. No one enjoys being told that they are wrong, especially in public or in an NIH CSR summary statement.

For experimentalists, it's a great risk to convert our pet theory to an explicit model; we might find it cannot explain even our own data, much less the data in our competitors' papers. Modelers face the same risk when we take the chance of comparing our model's predictions directly to experimental data. We may find to our horror that years hard work were for naught - that our model is hopelessly inadequate.

But if we don't take these risks, we are casting a no-confidence vote for modern science. We may sustain our careers at the cost of hampering real progress.

Wouldn't our science be more important and more fun if we ceased trying to protect ourselves from "a beautiful theory destroyed by an ugly fact."? Courage!

Non-binding recruitment

2008-12-22T08:24:00.000-08:00

I have a problem with recruitment.

The cell biological literature is replete with references to recruitment of proteins by other proteins. Typical is this one from the field of protein trafficking: "Arf recruits COP-I to the Golgi membrane."

At best this phrase introduces an unconscious synonym for binding; at worst it contains elements of vitalism or at least a Maxwellian demon. I've asked many biologists what they mean by recruitment, especially when they use binding and recruitment in the same paragraph. Responses vary but a summary might be: "Recruitment is definitely distinct from binding and I know recruitment when I see it."

So here is the challenge to readers of this post: Give us an example of recruitment that highlights the difference between recruitment and binding. Personally, I think this impossible.

How to measure the success of biomedical modeling

2008-12-18T08:59:00.000-08:00

Today's edition of the Predictive Biomedicine Newsletter includes an interesting piece from John Russel based on his attendance at IBM's invitation-only Modeling and Simulation Summit. Full disclosure: I was not invited.

Russell makes the provocative suggestion that perhaps what is needed to convince decision makers to embrace modeling is a central repository of modeling and simulation success stories. Many of us working in systems biology would agree with this recommendation. Many of us have told our students that this is the way to sell modeling to skeptical colleagues. But there is a problem and Russell adumbrates it in his next paragraph.

"Measuring modeling and simulation’s contribution to a project’s success is also a challenge. It’s often not easy to demonstrate that modeling was decisive versus incremental to a project’s overall success. Even when modeling is successful, other researchers on the project may feel they would have come to the correct answer soon enough without modeling."

The problem in a nutshell is how to define "decisive success" so that all the stakeholders agree on the successful/unsuccessful outcome at the end of the test.

To me what is missing is the control group.

We need expert teams aiming to solve the same problem with and without modeling. This is not easy to arrange. Moreover, we will still have the problem of how to agree that the problem is solved. And we won't know if one team is simply smarter or luckier than the other.

From another perspective, however, the experiment is already underway. Companies of all sizes whose business models are to sell the fruits of modeling and simulation are hard at work in biological, biomedical and pharmaceutical discovery. These are the people who provide the necessary tests. I'm emphatically NOT speaking about large research and development enterprises for whom an investment in modeling and simulation is mere bet hedging. I'm talking about those companies whose foundation is modeling. Companies like Entelos, and Merrimack and GNS who have bet the ranch on the conviction that human biology is far too complex to be manipulated successfully by the unaided human brain. Companies who understand that biology and medicine are HARDER problems than engineering a new airplane or a new chip or a new international communications network, or a new approach to managing financial markets or natural resources. Companies who recognize that no one can solve such problems without the aid of basic physical and chemical principles encoded in some sort of modeling framework. These companies have modeling built into their genomes. They absolutely KNOW that modeling is an essential element of biomedical research. These are the companies that will increase in value if we are right, and decrease in value if we are wrong. For the record, I myself have invested more than 40 years in applying modeling to biomedical research and 12 years in building small companies with modeling foundations. I believe.

What we cannot predict is which such companies will find a truly effective approach to leveraging modeling in biomedical and pharmaceutical research and development. Indeed, it may well be that a modeling division in an established pharmaceutical company will find one first.

But if you are looking for a list of modeling success stories, you could do a lot worse than tracking the fortunes of companies large and small, publicly held, venture-backed, or privately held, who are 100% committed to finding the most effective way to manage the modeling process in biomedical research. One of these, or perhaps one that is still just a great idea in someone's head is, without a doubt, going to make multiple billionaires AND a far healthier planet.

Slash and burn

2008-12-08T21:45:00.000-08:00

Slashes (/) appear frequently in the biological literature. But I'm worried we haven't come to a universal agreement on what they mean. Perhaps there already is a standard recommendation and some reader will take pity on me and post a URL, but I've asked a few biologists and heard multiple answers. I've also run a few Google searches and here is what I know at this point.

The most common use of / is to separate the individual items in a list of synonyms. For example: Rbx1/Roc1/Hrt1. Three names for the same protein.

Another usage is to specify a particular protein component in a multimeric complex. For example: SCF/Slimb represents a generic complex named SCF, which is made up of Skp1, Cul1, Rbx1, and one or another F-box protein. The notation SCF/Slimb indicates an SCF complex in which Slimb is the F-box protein.

And occasionally you see complexes specified using a slash as a ditto. Common examples are Arp2/3 and chlorophyll a/b.

I'm not a text miner but as a consultant I am often reading in a field of cell biology that is entirely new to me. Once I was led wildly astray by an author who named a complex by separating its components with slashes. In that case A/B/C meant a heterotrimer made up of three proteins: A, B, and C. This would be a text mining nightmare. Indeed, I found several bioinformatics papers whose sole purpose was to identify synonyms in biological text. Not an easy job.

I'm not sure what the forum should be, but it seems obvious to me that general agreement about the meaning of / would be worth promoting.

Personally, I'd like to see the colon (:) used as the separator of a complex's components. For example, The SCF complex mentioned above would be symbolized as:

Slimb:Skp1:Cul1:Rbx1. But that's another post.

What do YOU mean when you separate symbols with slashes?

The Desire for Simple Models

2008-11-28T18:02:00.001-08:00

If you spend much time among modelers you will occasionally find one for whom some models are "too complex." One simple measure of complexity would be the number of state variables involved in the model being discussed. Many years ago, my good friend Art Shoukas was giving a lecture on cardiovascular physiology and showed a slide of Arthur Guyton's famous model of the cardiovascular system. This model contains on the order of 100 state variables. Being a biomedical engineer, it was natural for Art to present this diagram, but what surprised me was that he pointed to Arthur's creation and said, "...this model is useless." Perhaps from the perspective of a first-year medical student, Art was right. The Guytonian diagram was indeed beyond comprehension as an integrated whole. But I remembered, magnifying glass in hand, working my way through that entire diagram when I was in grad school. To me it had (and still has) enormous power both didactic and scientific. So why is it that some people insist on simple models?

It may be simply that their pencil and paper analysis skills are better than mine. But I think there is something more fundamental at work here. I sense that they don't find a model satisfying unless they can "understand" it - by which I think they mean they can calculate its predictions without the need for simulation.

I was reminded of this today while reading Alfred Tauber's article, The Immune System and Its Ecology, in the April 2008 number of Philosophy of Science. Tauber has a keen eye for the evolution of modeling approaches in immunology while maintaining an agnostic position on which is best. My reverie began when he cited Levins and Lewontin (1985) who "... observed that the computer models of 25 years ago were not holistic, but rather only expressions of large scale reductionism." Surely they had the Guytonian model in mind.

Most human beings prefer models that explain everything and yet still fit inside a human brain where they can be manipulated successfully.

I've never felt this way and this is the point I wanted to make in this post. I think there are three reasons why "large scale reductionism" is the most useful approach for computational cell biology.

1) It proceeds by assembling the reductionist pieces adduced by individual scientists, without asserting that the model works in some holistic, known way. In effect, the assembled model serves as a working hypothesis that needs to be tested against many, many experimental data sets collected at all levels of physiological organization.

2) In general, we are not (at least I am not) smart enough to know how to simplify a complex model so that the simplification can serve as a trustworthy surrogate for the "full" system. Indeed, one of my teachers, Mones Berman, concluded that complex models should be built first and then simplified as it becomes clear what is essential and what is not.

3) It's vital for multi-scale models to retain the molecular details because it is at the molecular level that we most often intervene in our efforts to improve quality of life for the people who pay us to do biomedical research.

Maps involving post-translationally modified proteins

2008-08-22T14:18:00.000-07:00

Many software tools that create or process biological maps use a single protein name even if the protein undergoes one or more post-translational modifications. This makes sense if you want to link to UniProt or any other genomics-based protein database, because from the perspective of the genome there is one gene and one gene product.

But the concept of one gene - one protein died a long time ago. We clearly need to deal with a) multiple splice variants, b) alleles with different coding sequences, c) the possibility of many post-translational modifications (PTMs) and d) proteolytic cleavages of the initial translation product. Some tools (e.g. CellDesigner and the UCSD/Nature Molecule Pages) account for PTMs with annotations. CellDesigner even has an elegant graphical depiction of the PTMs associated with a given molecule. But typically the NAME associated with a post-translationally modified state is not changed. It's the name of the protein. This is fine if you are a CellDesigner user. The CellDesigner diagram clearly displays the different PTMs.

My worry is what happens when you export a model like this to other software tools using the SBML (or any other) standard. The problem is that if the standard does not capture the PTM annotations, then multiple states (species) in the resulting SBML model will have the same name even though they have different IDs.

To me it appears that best practices should require state (species) names to be unique. Ideally, the name will convey information about the PTMs so that recipients of exported models will know how states (species) differ from one another.

Perhaps someone who reads this post will know if the nascent proteomics databases intend to create multiple database entries for all the post-transcriptionally and post-translationally modified forms of a given protein.

In the meantime, if you are importing maps/models from other investigators, be sure to inquire how they have handled PTMs.

Hypotheses, Models and Maps

2008-08-20T21:14:00.000-07:00

Today's mail brought the August 8 number of Cell, possibly the most distinguished scientific journal on the planet. Certainly among cell biologists, it has no peer. When Cell turns to philosophy, you know something important is being debated at the highest levels of science.

David Glass, a distinguished Novartis physician-scientist and author of a recent CSHL Press book on experimental design, and Ned Hall, a Harvard (not MIT any longer?) philosopher who works on metaphysics and philosophy of science and specializes in the philosophical problems of quantum physics and analysis of causation, offer a Brief History of the Hypothesis with nods to all of my favorite philosophers except Kant and Whitehead. At issue, apparently, is whether it's time to replace Sir Karl Popper's Logic of Scientific Discovery with something more modern and visionary – something more like Francis Bacon's inductive Novum Organum (1620). They cite Thomas Kuhn's justly famous (and courageous, if you read his paper in Criticism and the Growth of Knowledge, delivered in a session with Popper as the chair) dismantling of Popperian logic and Robert Novick's frank dismissal of Popper as incoherent (meaning internally inconsistent). But to me all these attacks are like the Supreme Court ruling on narrow procedural grounds rather than interpreting the Constitution.

Popper's problem of demarcation asked, in effect, how we tell the difference between science and non-science. His answer was that a scientific theory (hypothesis) must be falsifiable by some conceivable experiment. Popper's view of scientific method, therefore, was called hypothetico-deductive. His views, now firmly ensconced in the public advice to prospective NIH grant applicants, tell us to start with the hypothesis and then propose experiments capable of testing it and perhaps falsifying it. His critics have always asked, where does this hypothesis come from? If you tell them it comes from scientific imagination, they pounce on you with gleeful screams of "imagination is just induction in disguise!" Probably they don't cite Kekule as an example.

I am no more than an amateur philosopher, but I think it's possible this centuries-long methodological debate has a resolution just like Kekule's answer for benzene - if you think of scientific method as a linear sequence of steps with an endpoint you quickly get into epistemological trouble, but if you see it as a loop you may be able to avoid this trap. Here is a map I've drawn to try to tame this problem for myself:

The map is from Nature Reviews Molecular Cell Biology (2001) 2:898-907 and a full description is available there.

When I first drew this, it was intended as a map of Popper's Logic. But students always wanted to know, just as Kuhn, and Glass and Hall would want to know, about the arrow labeled NO. They wanted to know HOW one goes about changing a hypothesis that fails to match the experimental data. I said, "Scientific imagination." They said, "That sounds like induction to me." And of course they were right. But to reject Popper's method because you uncover an anti-Popperian inductive step seems to me shortsighted. Moreover, the Glass and Hall proposal that we should abandon hypothesis building in favor of posing straightforward questions sounds naïve – especially so when their concluding example results only in a model of a genome sequence. If your question is something more aligned with the complexity of human biology, such as, “What is the cause of non-alcoholic steatohepatitis?” you are likely to need brilliant experimental designs and a very complex mechanistic hypothesis/model. I would have guessed this was true of cachexia too.

Scientists have a working model in their heads. Some of them call it a working hypothesis. All of them aim to test it and most of them, being human, have a vested interest in "proving" it correct. I believe each of them, however, could describe an experimental result that would convince them they were wrong, especially if it was their own experiment. This, to me, is Popper’s point. Kuhn is clearly correct that in the real world of modern science, no single experimental result is sufficient to unseat a reigning paradigm (hypothesis, model, theory). Glass and Hall are also clearly correct that working models are frequently reshaped (rather than rejected) by scientists aiming to account for new experimental results. Some of those scientists, however, would surely say they were revising their working hypothesis.

It does not seem vital to me to distinguish among the various terms scientists use to describe their current idea about how their system works. Diagram, map, hypothesis, model, theory, working hypothesis, working model, paradigm, cartoon, idea, computational model, computer model, mathematical model or mechanistic model – they all represent what we think we understand right now.

What does seem vital is finding better ways to generate models from large data sets and better ways to test them against even larger data sets.

There is a huge and growing chasm between “discovery science” and “hypothesis-testing science.” Chins are set and people in both camps seem certain they are right. This is not good for science. To me each side is half of the loop of scientific method. The new inductive “discovery science” (or systems biology) tools are the most powerful means we have to generate new hypotheses about complex biological systems, and the tools of modern computational cell biology (also called systems biology by some) are the most powerful means we have to test those complex hypotheses quantitatively.

But let there be no mistake. Experiments are pivotal. Both discovery science and computational cell biology would stop dead without great experiments, new experimental methods and the data they produce.

There is a natural and essential tendency to promote ones own scientific approach, and asking good questions is at the heart of what we do as scientists, but a proposal to abandon Popperian hypothesis testing just because inductive tools are beginning to work appears to treat the pursuit of knowledge as a zero-sum game in which Francis Bacon holds all the cards.

Glass and Hall quote Henri Poincare on the impossibility and undesirability of designing experiments without preconceived notions. They seem to imply that Poincare would side with them against Popper. But Poincare’s book is titled “Science and Hypothesis” and had it not been for Popper’s inflexible (and, to me at least, incomprehensible) stands on induction and probability, Poincare would have been, it seems to me, a Popperian hypothesis tester. After all, is was Poincare who said, “Science is made up of facts just as a house is built up of stones, but a collection of facts is no more a science than a heap of stones is a house.”

A general rule for cell biological map-making

2008-08-19T10:17:00.000-07:00

There are not many universal rules in map making. In fact, there are not many universal rules anywhere. You hear people say, "Never say never." They could as well say, "Never say always." But here is a rule that has stood the test of time, although it is rarely discussed and often violated:

When drawing a cell biological diagram (a signaling network, a biochemical reaction network, a genetic regulatory network, a membrane trafficking network, etc.) regulation must be represented as an influence of a molecule or a physical force on a process.

Diagrams (maps) showing molecules activating or inhibiting other molecules are often seen in the biological literature, but they are inherently ambiguous at just the moment when we want the author to be precise.

It can be reasonably argued that molecule-to-molecule regulatory diagrams are a necessary shorthand, but if you are interested in causality, many of us would urge you to adopt molecule-to-process regulatory diagrams. I know of at least three emerging standards that have adopted this principle, and you may find their sites educational:

SBML
BioPaX
SBGN Process Diagrams

Why SymbolsAndArrows?

2008-08-18T22:48:00.000-07:00

Biomedical research, like all creative pursuits, is map-making in an uncharted world. Maps fascinate. Maps teach. Maps lead and mis-lead. I remember once defining my life's overarching purpose as "making maps." A mapmaker must be part explorer, part adventurer, part mathematician, part artist, part engineer, part philosopher, part physicist, and part poet. I am some of these and also none of these, but I love maps of all kinds. I am a professional physiologist and biomedical engineer. I've been making bio-maps for more than 30 years - on paper, on blackboards and whiteboards, and on computer screens. I publish papers, but they must center on biology, not map-making. The symbols and arrows of hundreds of maps have led me here because I sense there is an important conversation to be had about maps, but I suspect the community of minds with whom I want to share this conversation is but a small intersection of a many-lobed Venn diagram. If you are fascinated by the big picture and passionate about the details, we may be on the same wavelength.