Scientific method: Defend the integrity of physics

Mike's Notes

This is part 1 of a series of 3 posts that reprint articles covering the crisis in particle physics that has emerged since scientists at CERN failed to discover a whole range of new particles as predicted by string theorists.

• Quanta Writers and Editors Discuss Trends in Science and Math (Quanta Magazine)
• A Fight for the Soul of Science (Quanta Magazine)
• Scientific Method: Defend the integrity of physics (Nature)

In my opinion, string theory is complete bollocks as is multi-universes. Nature is a book to be read, not written. Maths is essentially an invention rather than a discovery. Theory needs to be tested and proven by experimentation.

The article below is a stirring defence of science. I agree with it completely.

Resources

• https://www.nature.com/articles/516321a

References

• Reference

Repository

• Home > Ajabbi Research > Library > Subscriptions > Nature
• Home > Handbook > 

Last Updated

18/04/2025

Scientific method: Defend the integrity of physics

By: George Ellis & Joe Silk

On a Sandy Beach: 16/12/2014

Reprinted from Nature
Vol 516. Issue 7531

Attempts to exempt speculative theories of the Universe from experimental verification undermine science, argue George Ellis and Joe Silk.

This year, debates in physics circles took a worrying turn. Faced with difficulties in applying fundamental theories to the observed Universe, some researchers called for a change in how theoretical physics is done. They began to argue — explicitly — that if a theory is sufficiently elegant and explanatory, it need not be tested experimentally, breaking with centuries of philosophical tradition of defining scientific knowledge as empirical. We disagree. As the philosopher of science Karl Popper argued: a theory must be falsifiable to be scientific.

Chief among the 'elegance will suffice' advocates are some string theorists. Because string theory is supposedly the 'only game in town' capable of unifying the four fundamental forces, they believe that it must contain a grain of truth even though it relies on extra dimensions that we can never observe. Some cosmologists, too, are seeking to abandon experimental verification of grand hypotheses that invoke imperceptible domains such as the kaleidoscopic multiverse (comprising myriad universes), the 'many worlds' version of quantum reality (in which observations spawn parallel branches of reality) and pre-Big Bang concepts.

These unprovable hypotheses are quite different from those that relate directly to the real world and that are testable through observations — such as the standard model of particle physics and the existence of dark matter and dark energy. As we see it, theoretical physics risks becoming a no-man's-land between mathematics, physics and philosophy that does not truly meet the requirements of any.

The issue of testability has been lurking for a decade. String theory and multiverse theory have been criticized in popular books and articles, including some by one of us. In March, theorist Paul Steinhardt wrote in this journal that the theory of inflationary cosmology is no longer scientific because it is so flexible that it can accommodate any observational result. Theorist and philosopher Richard Dawid and cosmologist Sean Carroll have countered those criticisms with a philosophical case to weaken the testability requirement for fundamental physics.

We applaud the fact that Dawid, Carroll and other physicists have brought the problem out into the open. But the drastic step that they are advocating needs careful debate. This battle for the heart and soul of physics is opening up at a time when scientific results — in topics from climate change to the theory of evolution — are being questioned by some politicians and religious fundamentalists. Potential damage to public confidence in science and to the nature of fundamental physics needs to be contained by deeper dialogue between scientists and philosophers.

String theory

String theory is an elaborate proposal for how minuscule strings (one-dimensional space entities) and membranes (higher-dimensional extensions) existing in higher-dimensional spaces underlie all of physics. The higher dimensions are wound so tightly that they are too small to observe at energies accessible through collisions in any practicable future particle detector.

Some aspects of string theory can be tested experimentally in principle. For example, a hypothesized symmetry between fermions and bosons central to string theory — supersymmetry — predicts that each kind of particle has an as-yet-unseen partner. No such partners have yet been detected by the Large Hadron Collider at CERN, Europe's particle-physics laboratory near Geneva, Switzerland, limiting the range of energies at which supersymmetry might exist. If these partners continue to elude detection, then we may never know whether they exist. Proponents could always claim that the particles' masses are higher than the energies probed.

“The consequences of overclaiming the significance of certain theories are profound.”

Dawid argues that the veracity of string theory can be established through philosophical and probabilistic arguments about the research process. Citing Bayesian analysis, a statistical method for inferring the likelihood that an explanation fits a set of facts, Dawid equates confirmation with the increase of the probability that a theory is true or viable. But that increase of probability can be purely theoretical. Because “no-one has found a good alternative” and “theories without alternatives tended to be viable in the past”, he reasons that string theory should be taken to be valid.

In our opinion, this is moving the goalposts. Instead of belief in a scientific theory increasing when observational evidence arises to support it, he suggests that theoretical discoveries bolster belief. But conclusions arising logically from mathematics need not apply to the real world. Experiments have proved many beautiful and simple theories wrong, from the steady-state theory of cosmology to the SU(5) Grand Unified Theory of particle physics, which aimed to unify the electroweak force and the strong force. The idea that preconceived truths about the world can be inferred beyond established facts (inductivism) was overturned by Popper and other twentieth-century philosophers.

We cannot know that there are no alternative theories. We may not have found them yet. Or the premise might be wrong. There may be no need for an overarching theory of four fundamental forces and particles if gravity, an effect of space-time curvature, differs from the strong, weak and electromagnetic forces that govern particles. And with its many variants, string theory is not even well defined: in our view, it is a promissory note that there might be such a unified theory.

Many multiverses

The multiverse is motivated by a puzzle: why fundamental constants of nature, such as the fine-structure constant that characterizes the strength of electromagnetic interactions between particles and the cosmological constant associated with the acceleration of the expansion of the Universe, have values that lie in the small range that allows life to exist. Multiverse theory claims that there are billions of unobservable sister universes out there in which all possible values of these constants can occur. So somewhere there will be a bio-friendly universe like ours, however improbable that is.

Some physicists consider that the multiverse has no challenger as an explanation of many otherwise bizarre coincidences. The low value of the cosmological constant — known to be 120 factors of 10 smaller than the value predicted by quantum field theory — is difficult to explain, for instance.

Earlier this year, championing the multiverse and the many-worlds hypothesis, Carroll dismissed Popper's falsifiability criterion as a “blunt instrument” (see go.nature.com/nuj39z). He offered two other requirements: a scientific theory should be “definite” and “empirical”. By definite, Carroll means that the theory says “something clear and unambiguous about how reality functions”. By empirical, he agrees with the customary definition that a theory should be judged a success or failure by its ability to explain the data.

He argues that inaccessible domains can have a “dramatic effect” in our cosmic back-yard, explaining why the cosmological constant is so small in the part we see. But in multiverse theory, that explanation could be given no matter what astronomers observe. All possible combinations of cosmological parameters would exist somewhere, and the theory has many variables that can be tweaked. Other theories, such as unimodular gravity, a modified version of Einstein's general theory of relativity, can also explain why the cosmological constant is not huge.

Some people have devised forms of multiverse theory that are susceptible to tests: physicist Leonard Susskind's version can be falsified if negative spatial curvature of the Universe is ever demonstrated. But such a finding would prove nothing about the many other versions. Fundamentally, the multiverse explanation relies on string theory, which is as yet unverified, and on speculative mechanisms for realizing different physics in different sister universes. It is not, in our opinion, robust, let alone testable.

The many-worlds theory of quantum reality posed by physicist Hugh Everett is the ultimate quantum multiverse, where quantum probabilities affect the macroscopic. According to Everett, each of Schrödinger's famous cats, the dead and the live, poisoned or not in its closed box by random radioactive decays, is real in its own universe. Each time you make a choice, even one as mundane as whether to go left or right, an alternative universe pops out of the quantum vacuum to accommodate the other action.

Billions of universes — and of galaxies and copies of each of us — accumulate with no possibility of communication between them or of testing their reality. But if a duplicate self exists in every multiverse domain and there are infinitely many, which is the real 'me' that I experience now? Is any version of oneself preferred over any other? How could 'I' ever know what the 'true' nature of reality is if one self favours the multiverse and another does not?

In our view, cosmologists should heed mathematician David Hilbert's warning: although infinity is needed to complete mathematics, it occurs nowhere in the physical Universe.

Pass the test

We agree with theoretical physicist Sabine Hossenfelder: post-empirical science is an oxymoron (see go.nature.com/p3upwp and go.nature.com/68rijj). Theories such as quantum mechanics and relativity turned out well because they made predictions that survived testing. Yet numerous historical examples point to how, in the absence of adequate data, elegant and compelling ideas led researchers in the wrong direction, from Ptolemy's geocentric theories of the cosmos to Lord Kelvin's 'vortex theory' of the atom and Fred Hoyle's perpetual steady-state Universe.

The consequences of overclaiming the significance of certain theories are profound — the scientific method is at stake (see go.nature.com/hh7mm6). To state that a theory is so good that its existence supplants the need for data and testing in our opinion risks misleading students and the public as to how science should be done and could open the door for pseudoscientists to claim that their ideas meet similar requirements.

What to do about it? Physicists, philosophers and other scientists should hammer out a new narrative for the scientific method that can deal with the scope of modern physics. In our view, the issue boils down to clarifying one question: what potential observational or experimental evidence is there that would persuade you that the theory is wrong and lead you to abandoning it? If there is none, it is not a scientific theory.

Such a case must be made in formal philosophical terms. A conference should be convened next year to take the first steps. People from both sides of the testability debate must be involved.

In the meantime, journal editors and publishers could assign speculative work to other research categories — such as mathematical rather than physical cosmology — according to its potential testability. And the domination of some physics departments and institutes by such activities could be rethought.

The imprimatur of science should be awarded only to a theory that is testable. Only then can we defend science from attack.

Scientists Successfully Model a Living Cell with Software

Mike's Notes

This article I read in Scientific American led me to rebuild Pipi.

Resources

• https://www.scientificamerican.com/article/scientists-successfully-model-a-living-cell-with-software/

References

• Scientists Successfully Model a Living Cell with Software

Repository

• Home > Ajabbi Research > Library > Authors > Markus Covert
• Home > Handbook > 

Last Updated

11/05/2025

Scientists Successfully Model a Living Cell with Software

By: Marcus Covert

Scientific American: 01/01/2014

Reprinted from Scientific American, Jan 1 2014

In creating the first complete computer model of an entire single-celled organism, biologists are forging a powerful new kind of tool for illuminating how life works.

The crucial insight came to me as I leisurely rode my bike home from work. It was Valentine's Day, 2008. While I cruised along, my mind mulled over a problem that had been preoccupying me and others in my field for more than a decade. Was there some way to simulate life—including all the marvelous, mysterious and maddeningly complex biochemistry that makes it work-in software?

A working computer model of living cells, even if it were somewhat sketchy and not quite accurate, would be a fantastically useful tool. Research biologists could try out ideas for experiments before committing time and money to actually do them in the laboratory. Drug developers, for example, could accelerate their search for new antibiotics by homing in on molecules whose inhibition would most disrupt a bacterium. Bioengineers like myself could transplant and rewire the genes of virtual microorganisms to design modified strains having special traits—the ability to fluoresce when infected by a certain virus, say, or perhaps the power to extract hydrogen gas from petroleum—without the risks involved in altering real microbes. Eventually, if we can learn to make models sophisticated enough to simulate human cells, these tools could transform medical research by giving investigators a way to conduct studies that are currently impractical because many kinds of human cells cannot be cultured.

But all that seemed like a pipe dream without a practical way to untangle the web of interlinked chemical reactions and physical connections that make living cells tick. Many previous attempts, by my lab at Stanford University as well as others, had run into roadblocks; some had failed outright.

But as I pedaled slowly through the campus that winter evening, I thought about the work I had been doing recently to record images and video of single living cells. That's when it hit me-a way to make a realistic, functional simulator: choose one of the simplest single-celled microbes out there, a bacterium called Mycoplasma genitalium, and build a model of an individual germ. Limiting the simulation to just one cell would simplify the problem enough that we could, in principle, include every bit of biology known to occur in that cell—the unwinding of every rung of its twisted DNA ladder, the transcription of every message in that DNA into an RNA copy, the manufacture of every enzyme and other protein made from those RNA instructions, and the interactions among every one of those actors and many others, all building to cause the cell to grow and eventually divide into two "daughters." The simulation would generate, nearly from first principles, the entire drama of single-celled life.

Previous attempts had always tried to simulate a whole colony of cells because that is how almost all the data we have on cell behavior were collected: from populations, not individuals. Advances in both biotechnology and computing, however, had started to make single-cell studies much easier to do. Now, I realised, the tools were at hand to try a different approach.

Ideas whirred around in my head. The minute I reached home, I started sketching out plans for a simulator. The next morning, I began writing software code for just a couple of the many, many distinct processes that go on in a living microorganism. Within a week, I had completed several prototype modules, each one a software representation of a particular cellular process. The modules were producing output that looked fairly realistic.

I showed the work to a handful of other biologists. Most of them thought I was nuts. But I felt I was on to something, and two exceptional and daring graduate students, Jonathan R. Karr and Jayodita C. Sanghvi, saw enough potential in the approach that they agreed to work with me on the project.

Completing this model would mean creating dozens of such modules, combing through nearly 1,000 scientific articles for biochemical data, and then using those values to constrain and tweak thousands of parameters, such as how tightly enzymes bind to their target molecules and how often DNA-reading proteins bump one another off the double helix. I suspected that, even with the diligent help of collaborators and graduate students, the project would take years—but I also had a hunch that, at the end, it would work. There was no way to know for sure, except to try.

A Grand Challenge

As we set our sights on summiting this mountain, we took inspiration from the researchers who first dreamed of modeling life. In 1984 Harold Morowitz, then at Yale University, laid out the general route. He observed at the time that the simplest bacteria that biologists had been able to culture, the mycoplasmas, were a logical place to start. In addition to being very small and relatively simple, two species of Mycoplasma cause disease in humans: the sexually transmitted, parasitic germ M. genitalium, which thrives in the vaginal and urinary tracts, and M. pneumoniae, which can cause walking pneumonia. A model of either species could be quite medically useful, as well as a source of insight into basic biology.

The first step, Morowitz proposed, should be to sequence the genome of the selected microbe. J. Craig Venter and his colleagues at The Institute for Genome Research (TIGR)completed that task for M. genitalium in 1995; it has just 525 genes. (Human cells, in contrast, have more than 20,000.)

I was a graduate student in San Diego when, four years later, the TIGR team concluded that only 400 or so of those genes are essential to sustain life (as long as the microbes are grown in a rich culture medium). Venter and his co-workers went on to found Celera and race the federal government to sequence the human genome. They synthesised the essential genes of one Mycoplasma species and showed they functioned in a cell.

To me and other young biologists in the late 1990s, this gang was Led Zeppelin: iconoclastic, larger-than-life personalities playing music we had never heard before. Clyde Hutchinson, one of the biologists in Venter's band, said that the ultimate test of our understanding of simple cells would come when someone modeled one in a computer. You can build a functional cell in the lab by combining pieces without understanding every detail of how they fit together. The same is not true of software.

Morowitz, too, had called for building a cell simulator based on genome data from Mycoplasma. He argued that "every experiment that can be carried out in the lab can also be carried out on the computer. The extent to which these[experimental and simulation results] match measures the completeness of the paradigm of molecular biology"-our working theory of how the DNA and other biomolecules in the cell interact to yield life as we know it. As we put the puzzle together, in other words, it becomes more obvious which pieces and which interactions our theory is missing.

Although high-throughput sequencers and robotic lab equipment have greatly accelerated the search for the missing pieces, the floods of DNA sequences and gene activity patterns that they generate do not come with explanations for how the parts all fit together. The pioneering geneticist Sydney Brenner has called such work "low-input, high-throughput, no-output" biology because too often the experiments are not driven by hypotheses and yield disappointingly few insights about the larger systems that make life function-or malfunction.

This situation partly explains why, despite headlines regularly proclaiming the discovery of new genes associated with cancer, obesity or diabetes, cures for these diseases remain frustratingly elusive. It appears that cures will come only when we untangle the dozens or even hundreds off actors that interact, sometimes in unintuitive ways, to cause these illnesses.

The pioneers of cell modeling understood that simulations of whole cells that included all cellular components and their webs of interactions would be powerful tools for making sense of such jumbled, piecemeal data. By its nature, a whole-cell simulator would distill a comprehensive set of hypotheses about what is going on inside a cell into rigorous, mathematical algorithms. The cartoon like sketches one often sees in journal articles showing that factor X regulates gene Y ... somehow ... are not nearly precise enough for software. Programmers express these processes as equations—one of the simpler examples is Y = aX + b— even if they have to make educated guesses as to the values of variables such as a and b. This demand for precision ultimately reveals which laboratory experiments must be done to fill holes in knowledge of reaction rates and other quantities.

At the same time, it was clear that once models had been verified as accurate, they would take the place of some experiments, saving the costly "wet" work for questions not answerable by simulations alone. And simulated experiments that generated surprising results would help investigators to prioritise their research and increase the pace of scientific discovery. In fact, models offered such tempting tools for untangling cause and effect that, in 2001, Masaru Tomita of Keio University in Japan called whole-cell simulation "a grand challenge of the 21st century."

When still a graduate student, I was impressed by the early results of the leading cell modelers of the time [see box], and I became obsessed with this grand challenge. Even as I set up my own lab and focused on developing techniques for imaging single cells, the challenge remained in my thoughts. And then, on that February bicycle ride home, I saw a way to meet it.

Two Crucial Insights

It was clear that before we could simulate the life cycle of amicrobial species accurately enough to mimic its complex behaviors and make new discoveries in biology, we would have to solve three problems. First, we needed to encode all the functions that matter—from the flow of energy, nutrients and reaction products through the cell (that is, its metabolism), to the synthesis and decay of DNA, RNA and protein, to the activity of myriad enzymes—into mathematical formulas and software algorithms. Second, we had to come up with an overarching framework to integrate all these functions. The final problem was in many ways the hardest: to set upper and lower limits for each of the 1,700-odd parameters in the model so that they took on values that were biologically accurate—or at least in the right ballpark.

I understood that no matter how exhaustively we scrutinised the literature about M. genitalium and its close relations for those parameters (Karr, Sanghvi and I eventually spent two years culling data from some 900 papers), we would have to make do in some cases by making educated guesses or by using results from experiments on very different kinds of bacteria, such as Escherichia coli, to obtain certain numbers, such as how long RNA transcripts hang around in the cell, on average, before enzymes rip them apart to recycle their pieces. Without a way to constrain and check those guesses, we had no hope of success.

In that aha! moment in 2008, I had realised that modeling a single cell—rather than a bunch of cells, as almost all previous studies had done—could give us that constraint we needed. Consider growth and reproduction. A large population of cells grows incrementally; the birth or death of an individual cell does not change things much. But for a single cell, division is a very dramatic event. Before it splits in two, the organism has to double its mass—and not just its overall mass. The amounts of DNA, cell membrane and every kind of protein needed for survival must each double. If the scope of the model is constrained to a single cell, the computer can actually count and track every molecule during the entire life cycle. It can check whether all the numbers balance as one cell becomes two.

Moreover, a single cell reproduces at essentially a set pace.

M.genitalium, for example, typically divides every nine to 10 hours in a normal lab environment. It rarely takes fewer than six hours or more than 15. The requirement that the cell must duplicate all of its contents on this strict schedule would allow us to choose plausible ranges for many variables that would otherwise have been indeterminate, such as those that control when replication of the DNA begins.

I put together a team of physicists, biologists, modelers and even a former Google software engineer, and we discussed what mathematical approaches to use. Michael Shuler, a biomedical engineer at Cornell University who was a pioneer in cell simulation, had built impressive models from ordinary differential equations. Bernhard Palsson, under whom I studied in San Diego, had developed a powerful technique, called flux-balance analysis, that worked well for modeling metabolism. But others had shown that random chance is an important element in gene transcription, and cell division obviously involves a change in the geometry of the cell membrane; those other methods would not address these aspects. Even as a grad student, I had realised that no one technique could model all the functions of a cell; indeed, my dissertation had demonstrated a way to link two distinct mathematical approaches into a single simulator.

We decided, therefore, to create the whole-cell model as a collection of 28 distinct modules, each of which uses the algorithm that best suits the biological process and the degree of knowledge we have about it [see box ]. This strategy led to a patchwork collection of mathematical procedures, however. We needed to sew them all together somehow into one cohesive whole.

I thought back to an undergraduate course I had taken on chemical plant design. For the final class project, we used a powerful simulator package called HYSYS to sketch out a large refinery. HYSYS let us design each principal reaction to occur in a separate vessel. Pipes then connected the output of one vessel to the inputs of others. This framework connected many different kinds of chemical operations into an orderly, predictable system.

It occurred to me that this approach, with some modification, might work for our cell simulator if I was willing to make an important, simplifying assumption: that even though all these biological processes occur simultaneously in a living cell, their actions are effectively independent over periods of less than a second. If that assumption was sound, we could divide the life of the cell into one-second ticks of the clock and run each of the 28 modules, in order, for one tick before updating the pool of cell variables. The model would capture all the interconnectedness of biochemistry—the reliance of gene transcription and DNA synthesis on the energy and nucleotides produced by metabolism, for example—but only on timescales greater than one second.

We had no theoretical proof that this would work. It was a leap of faith.

While constructing our virtual cell, we put in software sensors to measure what was going on inside. Every run of the simulator, covering the entire life cycle of a single cell, churned out 500 megabytes of data. The numerical output flowed into a kind of instrument panel—a collection of dozens of charts and visualisations that, when printed, completely filled a binder.

The results were frustrating at first. For months, as we debugged the code, refined the math, and added more and better lab-derived constraints for the parameters, the cell refused to divide or behaved erratically. For a while it produced huge amounts of the amino acid alanine and very little else.

Then, one day, our cybernetic germ reached the end of its cell cycle and divided successfully. Even more exciting, the doubling time was around nine hours, just like that of living M. genitalium. Many other readings were still way off , but we felt then that success was within reach.

Months later I was at a two-day conference in Bethesda, Md., when I was called to the hotel's front desk between sessions.

"Dr. Covert? This package came for you."

Back in my room, I peeled open the box and pulled out a binder. As I spent the next hours flipping through hundreds of pages of plots and complex visualisations, my heart began to race. The great majority of the data looked just like one would expect from an actual growing cell. And there mainder was intriguing—unexpected but biologically plausible. That is when I knew we had reached the summit of that mountain that loomed so large years ago. The first computer model of an entire living organism was up and running. What would it teach us?

A Window into the Life of a Cell

After about a year of applying our new tool, we still see fascinating things every time we peer inside the workings of the virtual microorganism as it handles the millions of details involved in living and reproducing. We found, to our surprise, that proteins knock one another off the DNA shockingly often—about 30,000 times during every nine-hour life cycle. We also discovered that the microbe's remarkably stable doubling period is actually an emergent property that arises from the complex interplay between two distinct phases of replication, each of which independently varies wildly in duration. And the second-by-second records of the cell's behavior have allowed us to explain why it is that the cell stops dividing immediately when certain genes are disabled but reproduces another 10 times before dying when other essential genes are turned off. Those additional rounds of division can happen whenever the cell stockpiles more copies of the protein made from the gene than it needs in one lifetime—the extra is passed on to its descendants, which perish only when the store at last runs out. These initial results are exciting, but we may need years to understand everything that these simulations are telling us about how these microbes, and cells in general, function.

Our work with M. genitalium is only the first of many steps on the way to modeling human cells or tissues at the level of genes and molecules. The model that we have today is far from perfect, and mycoplasmas are about as simple as self-sustaining life-forms get. We have made all our simulations, source code, knowledge base, visualisation code and experimental data freely available online, and we and other investigators are already working to improve the simulator and extend it to a variety of organisms, such as E. coli and the yeast Saccharomyces cerevisiae, both of which are ubiquitous in academic and industrial labs.

In these species, the regulation of genes is much more complex, and the location within the cell at which events occur is far more important. When those issues have been addressed, I anticipate that the next target will be a mouse or human cell: most likely a cell, such as a macrophage (an attack cell in the immune system), that can be readily cultured and employed as a source of measurements to both tune and validate the model.

I cannot guess how far we are today from such technology. Compared with bacteria, human cells have many more compartments and exhibit far greater genetic control, much of which remains mysterious. Moreover, as team players within multicellular tissues, human cells interact more intimately with other cell types than bacteria do.

On February 13, 2008, I would have said that we were at least a decade away from the goal of modelling the simplest cell, and I would not have even considered attempting to model anything more complex. Now we can at least conceive of trying to simulate a human cell—if only to see how the software fails, which will illuminate the many things we still need to learn about our own cells. Even that would be a pretty big step.