Yves here. Given how significant clever marketing labels and propaganda have become, I am very leery of implicit Big Promise labels like Digital You and digital twin. It strongly connotes that medical AI will be able to create an informationally exact replica of your body.
This seems to run into what I call the Star Trek Transporter fallacy. The transporter appears to be widely agreed to be the most impossible future technology presented in the various series. There are many objections, such as the practical obstacles to “beaming” through all sorts of dense materials.
But I recall the early objections to the claims about how the transporter would work (that it would send not your particles but information about your particles, and then you’d somehow be recreated at the destination) was that it would take too much information, that it would be impossibly costly in energy terms to transport that much information, even if there was a way to identify and store all that information. I cannot find a rendition of that sort of complaint via Google, but the more recent ones are marvelous, even if not as on point with the Digital You concept. For instance, from Forbes:
The way the transporter allegedly works isn’t to move your actual atoms, but to teleport your information from one location to another, and to reconstruct you at your destination….
What you can do is transfer an arbitrary amount of information from one location to another through the process of quantum teleportation. The name is a bit of a misnomer, since this isn’t the teleportation of actual quantum particles, but of the information about the states of quantum particles. Make enough pairs of entangled particles between two different locations, and you can teleport that information from one location to another: you can move the state and the information of one object from point A to point B without having to move the object itself….
It’s possible that combining this technique with the emerging technology of quantum computing could enable the entire information encoding a living human being to be scanned in and teleported from one location to another. Or, if you didn’t see any need to destroy you, the original copy, perhaps you could be cloned entirely via this process!…
The challenge, however, is reconstructing that matter in the end state. Knowing what the information state of a human being is – including all their component particles – is one matter, but reconstructing that human being is quite another thing entirely. Despite a $14 trillion program launched by Russia – the National Technology Initiative – with the goal of teleporting a human being by 2035, it isn’t clear that this part of the technology is feasible given our current understanding of physics. To even dream of doing that would require not only putting all the particles that make you up back together in the same configuration, but with the same positions and momenta that they had before you were teleported. Think about the difference between a living human and a corpse of a human: there are no particles that are necessarily different, it’s simply the way those particles are positioned and moving in that configuration. But physics won’t even let you know those two pieces of information at the same time, much less reproduce them….
You see, there’s an inherent uncertainty between momentum and position for every particle, requiring that if you know one of those traits to a certain degree of precision, the other one becomes inherently uncertain so that the product of the two is always finite and non-zero. Lawrence Krauss, in his book The Physics of Star Trek, correctly identifies that one would need some type of hypothetical “Heisenberg Compensator” to account for this, which seems to violate the fundamental rules of quantum mechanics.
There are other obvious ways the Digital You overpromise fails. How can one possibly develop a training set for an individual? I engaged in oversharing re my orthopedic issues yesterday, where my stupidly attempting to implement advice for normal people got me injured. The reason I’m upset with myself is I should know better by now. The guy who did my hips perked up in agreement when I said, “My anomalies individually may not seem that serious, but in aggregate, they put me way out of band.” And orthopedics is simple compared to the rest of medicine. It’s mainly about mechanical systems.
I don’t mean to undermine KLG’s discussion and questions, which are more careful and nuanced than mine. And I’m not saying that there may not be merit in the approaches associated with this branding. I’m just very much put off by the hype.
By KLG, who has held research and academic positions in three US medical schools since 1995 and is currently Professor of Biochemistry and Associate Dean. He has performed and directed research on protein structure, function, and evolution; cell adhesion and motility; the mechanism of viral fusion proteins; and assembly of the vertebrate heart. He has served on national review panels of both public and private funding agencies, and his research and that of his students has been funded by the American Heart Association, American Cancer Society, and National Institutes of Health.
Artificial intelligence is the latest revolutionary development that will disrupt our world, said disruption considered without comment to be good for everyone and everything. Perhaps, but disruption – move fast (heedlessly) and break things – has had an uncertain past and an unknown future and its utility varies depending on whose lives are disrupted. For example: Deindustrialization, an obligate correlation of which has been the near-total loss of control of the real economy, despite a surge in paper profits leading to an income distribution mimicking that of the late-1920s. We all know how that turned out in a world that was not yet full. As for the future of the transnational corporation currently known as Meta, who knows? Hype rarely matches reality except sometimes in sports, where the scoreboard is unhidden and statisticians reign.
Artificial intelligence (AI) is naturally a topic of current conversation in the medicine. When ChatGPT dropped (the appropriate current use of that word) earlier this year, medical students were enthralled, especially the small cohort forever searching for the “magic fairy dust” that will provide the mythical short cut to the many competencies required of a physician. AI in medicine has been a long time coming, though, and this is covered well in the readable and optimistic Virtual You: How Building Your Digital Twin Will Revolutionize Medicine and Change Your Life (March 2023) by Peter Coveney and Roger Highfield.
There is a lot to consider in this book, and a complete review would require a companion commentary at nearly 1:1 scale, not too different from the subject of the (very) short story “On Exactitude in Science” (145 words) by Jorge Luis Borges, which is the epigraph of Chapter 1 . For now, there is no place to begin but the introduction, in which the authors write: “In the long term, virtual cells, organs and humans – along with populations of virtual humans – will help evolve the current generation of one-size-fits-all medicine into truly personalized medicine.” How this is to be accomplished is covered in ten succinct and engaging chapters. As outlined in the Chapters 1– 6, which are well focused and can stand alone, all we must do to create our Digital Twin  and Virtual You is:
- Harvest diverse data about the body
- Craft theory to make sense of all these data and use mathematics to understand the fundamental limits of simulations
- Harness computers to put the “spark of life” into mathematical understanding of the human body
- Blend insights of natural and artificial intelligence to interpret data and to shape our understanding
- Begin to build digital twin cells, organs, metabolic maps, bodies
- Stitch together different mathematical models of different physical processes that operate across different domains of space and time within the body
Regarding data, the requirements are truly “colossal.” The human body contains about 37 trillion cells (37e12, 37 followed by 12 zeroes), more or less. The number of different cell types runs into several hundred. These cells contain an uncountable number of molecules, but a reasonable guess is 10e26. The mathematical description of these data requires the use of differential equations, partial and ordinary, along with a complete description of essential boundary conditions. While this book is self-contained, at least one of these equations would have been nice to see without reference to the very good bibliography . How computers will handle these data is covered well overall, with potential limitations considered. The state of clinical data required – diffuse, incomplete, collected and stored in incompatible formats – is acknowledged.
Chapter 3 begins with Richard Feynman: “If we were to name the most powerful assumption of all, which leads one on and on and on to an attempt to understand life, it is that all things are made of atoms, and that everything that living things do can be understood in terms of the jiggling and wiggling of atoms (emphasis added).” I will defer to the authors on computer science and engineering:
The possibilities that beckon in this new era of computation are limited only by the imagination…the heavens will open up with the help of exascale machines,” which are, from the outstanding Glossary, “Computing systems capable of a million million million (10e18) floating point operations, or flops (floating point operations per second).
From this computer power comes AI and machine learning , and an “AI (that) is most powerful when working hand in glove with mechanistic understanding based on the laws of nature, where AI hypotheses can be tested in physics-based simulations and the results of physics-based methods are used to train AI. This is what we mean by Big AI” (Chapter 4).
Several examples include the identification of skin cancer lesions based on very well-trained AI (not included) and the analysis of histopathology slides (included). Image analysis is suited to machine learning because the training sets are large and largely complete. Another success for AI in biomedicine (i.e., biomedical sciences, medicine, Big Pharma) has been the prediction (actually retrodiction) of three-dimensional protein structures using AlphaFold based on a training set of nearly 200,000 known protein structures . Rational drug design depends on accurate structural data, and AlphaFold is often the necessary but seldom sufficient place to begin. As a happy aside, the published inferences of my first graduate student on the interaction of two proteins essential for cell motility have been largely confirmed by AlphaFold.
The use of Big Data and Big AI to deal with the “colossal dataset” consisting of more than 200 cell types among 37 trillion cells that vary in time and place during the lifespan of a human has led to intimations of the Virtual Cell (Chapter 6). In an early approach to modern synthetic cell biology, research by Craig Venter and associates on the small bacterium Mycoplasma genitalium and its cousin M. mycoides identified the core set of genes essential for (bacterial) life. Insertion of the M. mycoides chromosome into the “shell” of M. capricolum resulted in bacteria who had M. mycoides phenotype instead of M. capricolum upon replication. From Mycoplasma to the gut bacterium (and workhorse experimental model) Escherichia coli, which is a much more complicated free-living bacterium, is the next step in the creation of a virtual bacterial cell.
This research may lead to the rational design of microorganisms. It certainly emphasizes the importance of information in biological systems. In this case the substitution of one genome for that of a closely related small bacterium was remarkable but does not seem to be the advance it was touted to be. This analogy is somewhat strained, but the drivetrain of a 1968 Oldsmobile 98 would have functioned very well in the 1968 Buick Electra 225.
Information was a primary concern of theoretically minded biologists through the early days of the establishment of modern molecular biology, which consisted primarily of experimental elaboration of the Theory of the Gene. Information Theory will be an essential component of the Virtual Cell and with that, Virtual You. And with that would come the use of “logic modules” in the construction of Virtual You:
This kind of higher-level description would free us from having to understand all the chemical processes withing cells, just as a circuit diagram of resistors, transistors and so on frees us from having to know what all the electrons in an electrical circuit are actually doing. Or as Paul Nurse puts it, “We need to focus more on how information is managed in living systems and how this brings about higher-level biological phenomena”…The greatest hurdle (in the path of Virtual You) is developing the necessary theory to lay bare the emergent processes of life.
The Virtual Heart is covered in Chapter 7. The heart is probably the single organ most amenable to computational medicine because its function is primarily a problem of electrical and mechanical engineering and fluid dynamics. The components and boundary conditions are known as completely as possible. How the heart develops and works, from the electrophysiology of the pacemaker and the propagation of electrical signaling during cycles of contraction and relaxation to fluid dynamics, is well understood. The cellular basis of contraction is a mechanical problem mediated by well characterized anchor and contractile proteins of the cytoskeleton. This foundation can be used to understand the responses of the heart during recovery from injury, i.e., both adaptive and maladaptive remodeling of damaged tissue.
The Virtual Body is described in Chapter 8 and has arisen by taking the four basic steps leading to the creation of Virtual You: (1) gathering data, (2) developing theory, (3) judiciously using AI and (4) then working out “how to combine and blend theories that model the interactions of molecules in cells and tissues, which depends on all kinds of physics – mechanical, electrical, fluid flow, heat transfer and so on – that act over multiple scales, spatial and temporal.” The fifth and final step (5) is to “animate these multiscale and multiphysics models in computer simulations so that we can predict how Virtual You behaves in different circumstances.” Progress has been made on virtual muscles and skeletons, virtual lungs, virtual liver, virtual gut, and virtual metabolism. In consideration of the possibilities of a virtual brain, I will defer to two complementary reads that are 40 years apart in Julian Jaynes and Iain McGilchrist.
The “judicious” use of Big AI will be left undefined.
So, where does this leave us? It would be ridiculous to deny the advances in computational medicine since Alan Hodgkin and Andrew Huxley (yes, of that Huxley family) described the action potential in the squid giant axon and modeled the ionic currents through the plasma membrane using differential equations. We can now simulate metabolic flux through liver cells. Computational anatomy predicts where stresses might lead to bone fractures as a result of osteoporosis or metastatic lesions in weight-bearing bones. Radiation oncology has become a precision technique as both clinical particle accelerators and control software have improved. The ventilation of the respiratory tract has been described in detail, and this can be useful in modeling airborne infection, something very important these days and AI has been used in COVID-19 research (362,575 publications and counting).
Nevertheless, other considerations warrant discussion, primarily the role of theory in biology and medicine. The lament that biology is not a theoretical science is perhaps the central theme of this book. This is true. Theory in biology includes the Theory of Evolution and the Theory of the Gene and very little else. Various scientists have developed theories of the Origin of Life, from the Russian A.I. Oparin to Fred Hoyle, who wrote in the 1970s that life on earth came from outer space. It is not clear how current efforts will add to much here. Life happened and the interesting parts are what came after. In any case, there is no Standard Model of biology that includes a key component Leon Lederman called the “God Particle.”
The following gloss on theory in science is oversimplified and based on the views of a working biologist. In modern physics (and chemistry to a lesser extent) the standard progression has been Theory-Prediction-Practice. In biology the progression is Practice-Theory-Prediction-Practice. This obviously leads to complications for a medicine envisioned to have the Four P’s in the conception of Leroy Hood of the Institute for Systems Biology in Seattle: Predictive, Preventive, Personalized, and Participatory. As quoted, Hood speaks of “scientific wellness (that) leverages personal, dense, dynamic data clouds to quantify and define wellness and identify deviations from well state toward disease.”
Medicine, properly practiced, is already preventive, personalized, and participatory. Medicine is also a perpetual work in progress, and although medical practice has not always had these three attributes, these terms need no definition. That we lack a healthcare system that is preventive, personalized, and participatory as the default is a problem of political economy instead of biomedicine . Whether medicine can be predictive depends on the definition.
Prediction in science can be temporal or logical. Temporal prediction requires (near) complete knowledge of the state variables of a system. A signal example in Virtual You is the use powerful predictive models to predict the weather: “Optimism about the potential of digital twins in medicine is bolstered by our current ability to forecast weather.” True about the weather, but it is also true that over the past 100 years we have developed knowledge of the relevant state variables allowing us to predict the weather. Modeling these in real time has been very successful on fast computers. Yet sometimes it rains when the Weather Channel predicts clear skies. The state variables necessary to predict when a person will get which cancer or have a heart attack due to atherosclerosis are likely to remain unknown, however robust our Digital Twin and Virtual You become, courtesy of Big AI, Big Data, and Big Theory.
Logical prediction is another matter altogether and is the basis of prediction in biology and medicine. As noted in the Introduction:
Theory, that is, the mathematical representation of the laws of nature, plays a relatively diminished role in medicine and biology. Even the Darwin-Wallace  theory of evolution, regarded by some as the greatest scientific theory of all, does not admit of a mathematical description. That might sound shocking, but the reality is that, while basic predictions about the patterns of inheritance have been made since Gregor Mendel studied peas in the nineteenth century, the course of evolution in not possible to predict in any quantitative manner.
It is not necessarily true that a scientific theory requires mathematical representation, outside of the physical sciences. This passage also conflates the two related biological theories that do exist, the Theory of Evolution and the Theory of the Gene. The actual course of evolution cannot be predicted with anything approaching mathematical certainty, but that is not true for the inheritance of traits specified by genes.  What can be predicted is that when an experimental population, just as a natural population, is placed under selection pressure that population will evolve, for example, hereand more recently here. In medicine, it can be logically predicted that not everyone who smokes will get lung cancer but that 90% of those who do get lung cancer in any of its several kinds are or were smokers. Many of those who were not smokers were exposed to toxins in the air such as radon.
The lack of an overarching theory of biology that admits of a physics/engineering approach is seen as a near-fatal deficiency among those enamored with the theory of classical and modern physics. This physics envy has antecedents. In the early 20th century one of the leading experimental biologists of his time, Jacques Loeb, explicitly and militantly viewed biology as an engineering problem (Controlling Life: Jacques Loeb and the Engineering Ideal in Biology, 1987; the title is particularly apt and relevant today). Loeb’s view did not prevail, but his research on regeneration in marine invertebrates such as those found near the Marine Biological Laboratory in Woods Hole has been viewed as a foreshadowing of regenerative medicine (i.e., stem cell biology). Stem cells mediate regeneration in Hydractinia, but Loeb had no way of knowing that.
Not long after Loeb, J.H. Woodger published Axiomatic Biology (1937), which attempted to place biology upon a theoretical foundation like that of mathematics. His work was modeled on Principia Mathematica (1910-1913) by Alfred North Whitehead and Bertrand Russell. This is covered in The Life Organic: The Theoretical Biology Club and the Roots of Epigenetics (2016). Axiomatic biology is not useful.
Which brings us back to Virtual You. A major goal of AI in biomedicine is consilience, which is described in Virtual You simply as “a powerful unification of knowledge.” Consilience has been part of the background radiation of natural science since the certified polymath William Whewell  published The Philosophy of the Inductive Sciences in 1847. A popular exposition was published by E.O. Wilson in 1998, followed by a rejoinder from Wendell Berry in Life is a Miracle: An Essay Against Modern Superstition in 2001. Wilson and Berry, both giants in their respective fields, were at cross purposes but the argument is illuminating for those so inclined. Berry’s brief against the thoroughgoing reductionism that motivates many scientists is convincing.
While consilience is a noble goal, the theoretical unification of biomedicine is likely to be illusory if not superstitious. Physics is the scientific model for biomedicine throughout Virtual You, and this leads us back to the substance of the Feynman quote above: the most powerful assumption in the attempt to understand life is that all things are made of atoms, and everything living things do can be understood in terms of the jiggling and wiggling of atoms. This is an unwarranted assumption, which was Feynman’s point? As noted in Virtual You in reference to a seminal article(paywall) published by Alex B. Novikoff in 1945 , there is no privileged perspective from which to understand biology, and by extension medicine. From Novikoff:
Each level of organization possesses unique properties of structure and behavior…which appear only when these elements are combined in the new system. Knowledge of the laws of the lower level is necessary for a full understanding or the higher level; yet the unique properties of phenomena at the higher level cannot be predicted, a priori, from the laws of the lower level. The laws describing the unique (italics in original) properties of each level are qualitatively distinct, and their discovery requires methods of research and analysis appropriate to the particular level.
Getting back to the “jiggling and wiggling of atoms,” individual carbon, nitrogen, and phosphorous atoms have had no meaningful history since the first three minutes. This is not true of the biomolecules made of these atoms that are the stuff of every living organism. There is no theory or practice that allows us to progress from these atoms to the function of biological molecules and cells that have a common history of evolutionary change from the original common ancestor of all living things, more than 2.5 billion years ago. We are full of Higgs bosons, but there is nothing in them that can explain how an enzyme is regulated in the metabolism of lipids, proteins, or carbohydrates, with more from two papers I read in preparing this review.
The context-dependent, combinatorial logic of BMP signaling shows that signaling through bone morphogenetic protein is very complex at the cellular level and depends on near-inscrutable, perhaps random, conditions that probably cannot be known a priori. Another example from Virtual You is the molecular dynamics simulation of the interaction of RAS proteins with biological membranes. Yes, RAS is an oncogene commonly mutated in cancer. The key to understanding how RAS mutants lead to cancer is how they get stuck in a “feed-forward” mode, not necessarily what “100,000 microscopic simulations to model protein behavior over merely 200 milliseconds” might reveal. 
A different approach to a deeper understanding of biology at the cellular levels and beyond has been presented in a complementary read that I recommend highly to those interested in digging deeper: In Search of Cell History: The Evolution of Life’s Building Blocks by Franklin M. Harold. Harold has a gift, very similar to that to the authors of Virtual You, for explaining “how things work.” He is certainly one of the very few biologists to consider how “cell heredity transmits the living system’s pattern of global organization. Should that structure be lost, it cannot be reconstituted.”  Cell heredity also explains how the chromosome of one Mycoplasma can take over the cell of a related species and reproduce itself, which it could not do alone or in the shell of another unrelated bacterium.
Nor does Big Data seem likely to lead to the Big Theory that explains it all, even with the advent of quantum computing (Chapter 8). It is not clear how we might “craft theory to make sense of all these data and use mathematics to understand the fundamental limits of simulations,” as Big AI is used judiciously to go from atoms to molecules to cells to tissues to organs to individuals. A.B. Novikoff explained the emergent properties of biological systems 78 years ago. The authors of Virtual You also understand that “emergence” is a theoretical cliff that needs to be scaled. But the “logic modules” previously mentioned that would facilitate “this kind of higher-level description and free us from having to understand all the chemical processes within cells, just as a circuit diagram of resistors, transistors and so on frees us from having to know what all the electrons in an electrical circuit are actually doing” remain obscure.
I am no physicist, but I did take physics and if I remember correctly, we do know what the electrons in an electrical circuit are doing. And that their ultimate “purpose” depends on the organization of the circuits, not the flow of the electrons. I am a biologist, however, and I do know that for medicine to be practiced properly, reliance on black boxes as explanatory tools in the understanding of normal and abnormal physiology is likely to fail at some point. Inputs must be understood as well as outputs, and how the latter appeared. The past three years have reminded of this in the strongest terms.
Where do we go from here? According to Virtual You:
In future, medicine will increasingly be led by scientific insights into health and treatments that, akin to engineering, are based on theory, data, modelling, and insights about how your own body works. Rather than always looking backward at the results of clinical trials, medicine will become truly predictive…we can speed up diagnosis by giving a regular update on the physiological state of a person (using) wearable devices and smart phones. Machine learning can be trained to look for warning signs of health problems.
All well and good, but the theory behind this engineering must be unlike anything that seems possible. As for the accumulation of the required data, in the age of Surveillance Capitalism it is unlikely to be used as the advocates of Virtual You (mostly) assume it will be used. This is considered in the final Chapter 10: Healthcasts to Posthuman Futures, as it should be. But the conclusions are worrisome:
Virtual You will further blur the boundaries between human and machine…Yet virtual twins will transform the human condition by increasing confidence in methods that we can use to enhance human intellect and physiology, whether by implants, drugs or gene editing…Your twin will offer doctors a risk-free means to test experimental treatments…As it becomes easier to augment our mind and body, just as an engineer can turbocharge a car, the concept of what it means to be human could gradually change…virtual humans will enable us to plan for a posthuman future…the Janus-like nature if technology has been apparent for more than a million years: ever since we harnessed fire, we knew we could use it to stay warm and cook but also to burn down our neighbors’ houses and fields.
Engineering, again, and the use of Prometheus frankly clanks. Nor is the “posthuman future” particularly interesting. There is no doubt that computational medicine will lead to incremental advances at the margin of biomedicine. Computational cardiology already has. The biologist and physician are naturally enthusiastic about incremental advances. They are generally the only kind of advance in biomedicine, and some of them will be revolutionary. We seldom know which ahead of time, though. Rather than depending on Big AI to build our Digital Twin and Virtual You based on theories that come out of Big Data in and of itself, we could transform healthcare so that it is truly universal, focused on the preventive, personalized, and participatory. Medicine, properly practiced today, is not one-size-fits-all nor does it have to be, whatever the fate of our digital twin and virtual you. The three P’s will be enough, and “predictive” will be useful at the far margin where logical and temporal prediction coincide.
Still, we can look forward to the judiciously managed revolution in biomedicine that is expansively covered in this remarkable book. Keeping up with the development of our digital twins will be interesting. I will not live to meet mine, but the possibilities are intriguing. They are, however, constrained by more than our collective imagination, for reasons scientific, theoretical, political, and cultural.
In the meantime, we should remember the words of T.S. Eliot that are relevant to the practice of medicine as a healing science and art:
All our knowledge brings us near to our ignorance,
All our ignorance brings us nearer to death,
But nearness to death, no nearer to God.
Where is the Life we have lost in living?
Where is the wisdom we have lost in knowledge?
Where is the knowledge we have lost in information?
The cycles of Heaven in twenty centuries
Bring us farther from God and nearer to the Dust.
The Rock, 1934
 A digital twin can be viewed as a useful simulacrum of an actually existing object or system and can be used to simulate engineering design features and changes in supply chains and factories. The goal here is to effectively extend this engineering concept to human biology and medicine.
 I used an early edition of this textbook while totally outclassed in a room full of math, chemistry, and physics majors. We didn’t model anything that wasn’t a simple engineering problem, but the only computer that was (not so) available to an undergraduate was an IBM 370 mainframe that required punch cards.
 From the Glossary, which is outstanding: Machine learning is an inference-based approach to finding relationships between input and output data withing digital computers by an algorithmic process known as “training” in which correlations are detected and honed through iterative adjustment of large numbers free of parameters. Often used interchangeably with artificial intelligence. Examples par excellence are artificial neural networks. Deep learning is a term used to describe such learning in artificial neural networks that have more than three layers.
 To get deeper into the weeds, the history of the protein folding problem is long and complex beginning with Levinthal’s Paradox. This Wikipedia entry covers the history of AlphaFold well, based on discussions among experts that I attended when I was a postdoctoral fellow. This regularly scheduled Clash of Titans was often as entertaining as it was illuminating.
 And it is becoming clear that these “data clouds” that exist in server farms somewhere out there are not used to improve healthcare. Rather, along with electronic medical records they are used to fine tune the extractive operations of an extremely lucrative for-profit biomedicine.
 Alfred Russel Wallace has not been ignored but he has also not been as appreciated as he should have been. Although he died 110 years ago, he was as much of a scientist for all time, including our time, as any other: Radical by Nature: The Revolutionary Life of Alfred Russel Wallace by James T. Costa.
 Gregor Mendel has been accused of reporting results that were “too good to be true,” but it should be noted that the discipline of conventional frequentist statistics did not exist when he did his experiments on peas in the monastery garden in Brno. As his work developed, he did have expectations, and a pea with one crease or that was yellow-green could be counted in whichever category that made the most sense to him, wrinkled or yellow or green. All scientists must guard against this at all times.
 From Wikipedia: One of Whewell’s greatest gifts to science was his word-smithing. He corresponded with many in his field and helped them come up with neologisms for their discoveries. Whewell coined, among other terms, scientist, physicist, linguistics, consilience, catastrophism, uniformitarianism, and astigmatism; he suggested to Michael Faraday the terms electrode, ion, dielectric, anode, and cathode.
 A.B. Novikoff was a pioneering cell biologist before the field had a name. I included this reference in a research proposal to a public funding agency and one reviewer objected vehemently because (paraphrase) “absolutely nothing from 1945 could be relevant in 2015.” Seems funny now but it wasn’t in real time.
 Paul Nurse, p. 8-9: Paul Nurse, director of the Francis Crick Institute in London and former assistant editor of the Journal of Theoretical Biology (est. 1961), told us how he was weary of reading papers that used clever technology to make measurements that come to “barely any significant conclusions.” In my field this would be the measurement of the “pico-newton (1e-12) slip bond between two proteins involved in cell motility,” a technical tour de force with very little significant meaning.
 This reminds one of the protein folding problem from Note 5. Christian Anfinsen was awarded the Nobel Prize in Chemistry for demonstrating that the amino acid sequence of a protein determines its final active structure. But his work also showed this to be true only when the experiment with the small enzyme RNase was conducted in an environment that mimics the interior of the cell (reducing rather than oxidizing). Thus, while the amino acid sequence is sufficient, cell heredity is essential for proper folding