Orac Lies Again. Part II
In part I, I discussed yet again what the word predict means in science and how it can be used in two different ways. In this blog, I will address the specific use of mice to predict human response as Dr Gorski discusses in his essay. Mouse avatars are mice that have been implanted with human cancers and are treated simultaneously with the patients from whom the cancers have been removed. The thinking being that if the treatment does or does not work in the patient, the mouse can be studied in an attempt to explain why as well as functioning as a test subject for potential future treatments. This is a good example of using animals as predictive models for human response to drugs. Indeed, the NY Times article discussed by Dr Gorski, Seeking Cures, Patients Enlist Mice Stand-Ins quotes Colin Collins, a professor at the University of British Columbia, as saying: “The mice allow you the opportunity to test drugs to find out which ones will be efficacious without exposing the patient to toxicity.” That's pretty straightforward.
Dr Gorski analyzes the concept and the current implementation of the concept and concludes: “I do think that sequencing cancer genomes and doing expression profiling, then using mouse models like this, could hold considerable promise for predicting individual patient response to different regimens.” In fairness, Dr Gorski was critical of many aspects of the practice, but as the above indicates, he thinks it has potential. Dr Gorski continues: “So in the end, what we have here is an animal model that very well might be predictive of human response in a way that is more direct and individualized.” Again, this is stated with caveats, but his position is clearly that such models might prove to be predictive modalities. I will now explain where and why I disagree with Dr Gorski’s position.
After someone understands the prediction problem in animal modelling, based on a survey of the empirical evidence from many fields, invariably he will grant that animal models so far have failed as predictive modalities but insist that this does not mean they will always fail. And you know what? He is right!
If one looks at a great amount of empirical evidence from fields as diverse as toxicity testing, bioavailability testing, cancer research, HIV/AIDS research, stroke research, and so forth one can make a definitive case for the fact that animal models have failed to function as predictive modalities for human response to drugs and disease, the occasional correlation notwithstanding. The phrase “for human response to drugs and disease” is key here, as I have pointed out many times that animal models can be predictive for response to perturbations that occur at lower levels of organization, where complex systems can be described in terms of simple systems. For example, tossing a monkey out of a plane at 10,000 feet above ground level will be predictive for the outcome for tossing a human out at that altitude. Splat! That is not the same level of organization in a complex system as where drug and disease response occurs. For more, see Animal models and conserved processes. (An understanding of complexity science is essential to understanding my position. If you cannot list at least seven characteristics of a complex system, you don’t know what you are talking about and need to put in some time studying before commenting on all this.)
So the empirical evidence only gets you so far when debunking the animal model as a predictive modality. Just because it has not worked up until now does not mean it will never work. Maybe we can tweak the system and make an animal model that will predict human response to drugs and disease. Historically, anti-vivisectionists cited examples where animals and human demonstrated very different responses to the same drug or disease and claimed the animal model per se was not doing what scientists claimed it was doing. Scientists, at least the more honest ones, acknowledged that such was indeed the case but promised that further research would allow them to invent a model that functioned as a predictive modality and hence their research efforts should continue to be funded. After all, if they ever did come up with a model that had a high PPV and NPV for say teratogenicity or carcinogenicity, all the failures would be worth it. While some may not buy that line of reasoning there is nothing obviously wrong with it from a science or logic perspective.
I now must digress and discuss some basic principles in philosophy of science.
Nonscientists, and even scientists, usually do not understand the definitions of very basic concepts of science. For example, theory, law, and hypothesis are routinely confused. Williams  discovered that, of the graduates in science that Williams surveyed:
- 76% equated a fact with 'truth' and 'proven'
- 23% defined a theory as 'unproven ideas' with less than half (47%) recognizing a theory as a well evidenced exposition of a natural phenomenon
- 34% defined a law as a rule not to be broken, and forty-one percent defined it as an idea that science fully supports.
- Definitions of 'hypothesis' were the most consistent, with 61% recognizing the predictive, testable nature of hypotheses.
Williams states that the students did not understand the differences between laws, theories, and facts and further did not appreciate the difference between a scientific theory and hypothesis. Some thought hypothesis and theory were the same thing.  As these definitions are relevant to my position I will pursue the topic for a few more paragraphs. The National Academy of Sciences (USA), explains theory as follows:
In everyday usage, “theory” often refers to a hunch or a speculation. When people say, “I have a theory about why that happened,” they are often drawing a conclusion based on fragmentary or inconclusive evidence. The formal scientific definition of theory is quite different from the everyday meaning of the word. It refers to a comprehensive explanation of some aspect of nature that is supported by a vast body of evidence. Many scientific theories are so well established that no new evidence is likely to alter them substantially. . . . One of the most useful properties of scientific theories is that they can be used to make predictions about natural events or phenomena that have not yet been observed. [ p11]
Examples of theories in science include:
- The Big Bang Theory
- Cell Theory
- The Theory of Evolution
- Atomic Theory
- Kinetic Theory of Gases
- The Germ Theory
- Chaos Theory
- Theory of Special Relativity
- Theory of General Relativity
Before defining hypothesis I should contrast and compare laws of science with theories. Laws are similar to theories in that both can be used to predict outcomes and both have dramatic quantities and quality of evidence to support them. Laws tend to be confined to specific situations while theories usually include explanations in addition to predictions. Theories tend to answer how and why questions whereas laws simply predict outcomes or behavior. Laws of science include:
- Newton’s three laws of motion
- Boyle’s law
- Law of conservation of energy
- Joule’s first and second law
- The four laws of thermodynamics
Hypothesis on the other hand, refers to ideas for explaining natural phenomena that have not been tested or that have been inadequately tested. Hypotheses are works in progress. They may turn out to be true or they may not. Hypotheses may come from observations, experiments, or simply from thinking about a problem. One should say: “I have a hypothesis that X is correlated to Y;” not: “I have a theory that X is correlated to Y.”
So, why is the above important to this examination of mouse avatars?
Whereas anti-vivisectionists had historically criticized animal models on ethical grounds and scientifically based on examples, Niall Shanks and Hugh LaFollette began writing in the early 1990s about more general concerns based on the theory of evolution and chaos and complexity theory. Shanks and I expanded on this in our books and articles. I would not suggest that Shanks et al came up with a new theory in science. I would say that we combined parts of two existing theories and discovered why animal models will always fail as predictive modalities at the level of organization relevant to drug and disease response. Whether one wishes to classify this as a new scientific theory, “Modelling evolved complex systems theory,” (which I do not) or just an application of two old theories—complexity and evolution—is not relevant to this discussion. The point is that scientific theory regarding animal models does exist and it does exactly what a scientific theory is supposed to do: make sense of findings from disparate fields and “make predictions about natural events or phenomena that have not yet been observed.”
The fact that animals and humans are examples of evolved complex systems means that merely tweaking the animal model will not be sufficient to make it into a predictive modality. Systems that are complex are by definition more than the sum of their parts and hence tweaking will produce something that is still differently complex from humans. The contribution of Shanks et al is placing the empirical evidence, which is abundant, in the context of the theory pertaining to evolved, complex systems.
If an animal model was invented tomorrow that showed a high PPV and NPV for a disease or drug, evolved complex systems theory predicts that the level of organization where the response is acting is either low enough to be considered a simple system or occurring in a module that is the exception to the rule. In other words the module is simulable even though it is a complex system. Or it could be just random luck—coincidence. This means that if such a model were invented it would not be a violation of a scientific law but rather the exception to a scientific theory. Exceptions to theories are very rare (some germs cause disease in one person but not another) and when found are usually explained by the presence of other factors—perturbations occurring at lower levels of organization, for example. Spending money on research premised on finding exceptions to theories in biomedical science is a fool’s errand. It also costs lives because that money could fund other research projects.
Mouse avatars are yet another example of the animal model community admitting that animal models are not predictive yet, but that this one might be the answer. They are asking society to continue to fund the game, as this one just might be the winner. “Step right up, young man and win one for the little lady. Today could be your lucky day!” In reality, the game is rigged.
1. Williams J: What Makes Science 'Science'? The Scientist 2008, 22:29.
2. Committee on Revising Science and Creationism: Science, Evolution, and Creationism. Washington DC: National Academy of Sciences; 2008.