Some Thoughts on Cancer from a Hacker

Just got through reading “The Emperor of All Maladies” on my Kindle today. Both my mom and dad died from Cancer, so it seemed like a good idea to read up on the subject.

This was an excellent book which provided a great historical overview of Cancer and how science has come to grapple with it. At one point the author makes the joke that Cancer has the Kevin Bacon thing going on: take any subject, and within 7 logical jumps you’ll be at Cancer research. It touches on almost all areas of modern life and science.

If you’ve suffered from Cancer or know somebody who has, this is a good book to read. A bit long and detailed, but it’s a big subject.

Now that I’m done, it got me to thinking: the structure of how we study Cancer might be entirely inappropriate to the disease itself.

Not to spoil the book (we don’t cure Cancer in the end), but as it turns out current belief is that each Cancer is a highly individualized thing — it’s your body’s programming breaking down, and each person’s programming breaks down in it’s own way. Yes, there are broad generalities — leukemia looks like leukemia — but even then, there are lots of different subtypes, dozens or hundreds of mutations can lead to the disease, and only a few of those might be critical. Indeed, it looks like part of the discussion nowadays is whether there’s some way of simplifying Cancer in order to create and provide treatment.

In response to this terrible uniqueness of the disease, I think, several new sciences around bioinformatics have sprung up. There’s a school of thought that says that what we need is not a cure: what we need is a computerized system that “debugs” what’s wrong with each person’s cancer and provides unique therapy appropriate to that. Lots of five-dollar words here, including proteomics, cytogenic chemotherpathy, etc.. The water’s too deep for me, but I get the part about complex systems of programming breaking down. Been there, done that.

This analogy makes a lot of sense to me as a layman, but I wonder about the way governments and scientists discover and approve new drugs. Does the solution fit the system we’ve created to find the solution?

The entire idea of empirical science is that we gather data, note patterns, make hypotheses, then prove them. Mathematical induction is key part of it: if it works a few times and can be reproduced, it will work always. But Cancer is exactly the opposite of that. In fact, one of the reasons it’s been so tough to handle is our idea that it’s just one disease we’re facing. It’s not.

So let’s suppose that there are a million different proteins which either are under or over expressed due to a Cancer. _If_ you had ways of correcting each one, and _if_ a computer could do the analysis, theoretically you could create an intervention that would be tailored to that particular cancer, changing and evolving as the “bugs in the programming” for that particular person mutated.

But how could you test this in a double-blind study? Would you test each one of the two million interventions? Impossible. Test the computer program? Non-conclusive (and perhaps not able of being reproduced, as the program should be evolving along with technology).

I don’t see any way the government could approve of such a therapy, and more importantly, I don’t see any way private companies could make money from helping folks, unless you let them start patenting each one of the two million interventions, which would effectively bring research to a standstill.

I think the way we do science, the way we do medicine, the way we regulate markets where companies can help each other — all of these things don’t match up to the solution we’re working towards in Cancer research. And a lot of people are going to die because of it.

Here’s hoping I’m wrong.

Share
If you've read this far and you're interested in Agile, you should take my No-frills Agile Tune-up Email Course, and follow me on Twitter.

17 thoughts on “Some Thoughts on Cancer from a Hacker

  1. Marty Williams

    I’m not a doctor or a geneticist, but looking at this from a computer viewpoint makes me want to figure out a way to restore a backup of the programming to the effected area. Run a gene analysis on the cancer, compare to a good copy from unaffected tissue, do a compare and figure out the difference, create a good “gene load” and place it in a retrovirus, inject into the cancer.
    I’m sure everything is harder than it looks, but when my data goes wacky, restore.

    Reply
  2. DanielBMarkham

    I’m with you, Marty.
    What it looks like today is that when we release a new “cure” today, all we’re really doing is addressing the protein problems when it’s only one or two proteins. Each of those drugs takes years to develop, and each only hits one protein.
    With over a million proteins, and god knows how many combinations, simply “curing” all of that would be an incredible feat that’s damn near impossible. But assuming you even got that far, as you point out, all you’re doing is treating the resulting proteins, not the “cause” of the bug. These are drugs you take for the rest of your life.
    To “fix” the problem, as you say, you’d either need a backup or a way of computationally fixing the broken parts of the sequence. Then you’d need to find something — a retro virus, perhaps? — that would go in and fix it for all the cells. It would be like creating your own “anti-cancer” for each person that would uniquely fight the type of cancer that person had.
    No doubt about it, the whole problem looks almost intractable, sad to say.

    Reply
  3. Jonathan Rosen

    Interesting discussion. It is certainly true that our current method of curing cancer is kludgey at best (basically, the current idea is to use drugs that kill everything in the affected area, and since the cancer cells are a relatively small subset of the overall cell population, your normal cells can come back, and the cancer cells will be gone).
    In the future, there will almost certainly be much more targeted treatments. The problem with retroviral treatments is that they are difficult to control/target, and the potential for escape mutants from the retrovirus itself could be highly problematic. You could end up creating some kind of supervirus that actually gives people cancer. With that being said, we may eventually get to a place where we can program retrovirus’ to do our bidding. However, we aren’t anywhere near that point right now.
    The first step is really with better bioinformatics, which you hinted at. Genetic sequencing is getting faster/cheaper, and once we are able to sequence an entire human genome quickly and on the cheap, we may be able to isolate cancer cell genetics as you have implied, identify the broken gene, and find a way to target those broken cells specifically.

    Reply
  4. DanielBMarkham

    The hidden assumption here should probably be brought out: that we finally understand what causes the millions of diseases that we call “cancer” Gene mutations into oncogenes due to chemicals, viruses, or simple mutations. These oncogenes either are turned on or off (or the genome otherwise mutated) in various combinations to create runaway conditions.
    We’ve made this mistake over and over again — assuming that we have some idea of the mechanism that we can attack. We’re nowhere near thinking that bad humours cause cancer like they did 500 years ago, but a little humility is probably in order. Already I’m hearing in the press that the human genome project may not be all that it was cracked up to be — that there may be lots of “missing” key information in material in the cell that’s not in the genome. Like I said, the water is too deep for me, but I’d just be careful we don’t get cocky. Even if things work the general way we think they work — and the evidence is looking better all the time — protein folding and customized genetics and making cancer to fight cancer? This is all stuff that was science fiction just a few years ago. We’ve a long way to go to any kind of generalized solution, even if we can now imagine one.
    I’m just happy that it looks like we’re slowly moving away from high doses of poison as a way to cure folks. One quote from the book seems apt “It’s like trying to cure a dog of fleas by beating him with a stick”

    Reply
  5. John

    Despite throwing in the word “hypotheses”, your description of how science works is wrong (as Karl Popper showed): induction, as you define it, is never involved.
    However, I think you are correct that there is a misfit between the practice and methodology of medical trials and attempts to make radical improvements to medical treatment.
    Here’s an interview about one such crazy idea: WILT. We need more crazy ideas in medicine, because they are the ones that will lead to big improvements.
    http://newcancermentality.blogspot.com/2011/01/new-cancer-mentality-interview-with_27.html

    Reply
  6. Richard Rodger

    The key question with cancer is why do random mutations always lead to the same patterns of runaway growth. I read about a really great idea for explaining this a few months ago, but I have the lost the reference.
    The basic idea is that our DNA contains ancient survival and propagation mechanisms more appropriate to slime moulds and single cell animals. Over 100′s of millions of years, a complex overlay of genetic controls has developed to keep these mechanisms in check, so that multi-cellular organisms can exist and function.
    But this complex overlay is fragile – due to it’s complexity. When it goes wrong, the ancient mechanisms reactivate. It is *not* the individual unique mutations that produce cancer directly.
    So there is hope for treatments that will work for many people. If this idea is correct, it is not the case that only individualized treatments are required.

    Reply
  7. caf

    If you came up with such a therapy, it could simply be tested for safety and efficiacy as a complete therapy. The fact that the underlying mechanism works differently for different people wouldn’t matter – as long as double-blind studies showed to be safe and to work, that’d be enough. Medical trials aren’t like using a mathematical proof to prove the correctness of a computer program – they’re more like a Monte Carlo black box test.

    Reply
  8. TerraHertz

    Here are a few more things to look into regarding cancer:
    * Telomeres, telomerase & cancer. Two things must go wrong in a cell to produce cancer. First, something with the cell’s signaling and reproduction control systems. AND, secondly and much more specific, the gene for production of telomerase must get turned on, thus making that cell line immortal. We all get ‘cancer’ every day, where some cell goes into uncontrolled replication. But without telomerase those cell lines die out after a couple of score divisions when they hit the ‘telomere fuse’ limit. Which happens before that cancerous cell mass even gets to pinhead size.
    Telomere fuses and the related control genes are also very interesting from the perspective of mortality and life extension research.
    * Vitamin B17, aka Laetrile. A naturally occurring substance that many claim very effectively targets and kills cancerous cells. The common story: ‘can’t patent it so not profitable, works well, therefore suppressed by medical establishment since it would cut their profits on other treatments.’ Judge for yourself. My only practical comment is that I’ve tried eating apricot pits, and found the stories I was told as a child about them being deadly due to containing cyanide are in fact bullshit – as the B17 proponents claim.
    Some links here: http://everist.org/archives/links/!_B17_Laetrile_links.txt
    Unfortunately I only learned of the B17 story _after_ my dad and best friend died of cancer, so didn’t have an opportunity to test if there’s any truth to the claims.

    Reply
  9. Ryan Marsh

    xThis is a great article and while Daniel may be on to something, the sad truth is: It doesn’t f***ing matter. Here’s why:
    When my wife and I lost our two year old daughter to cancer we started a nonprofit to raise money “for research” we thought. What we found when we got very prominent doctors behind closed doors is a system of bureaucracy that has led to the ineffective local maximums in research that have left heavily funded diseases such as Neuroblastoma (MYC-N amplified) with an increase in survival rate that amount to a rounding error.
    I’ll keep this short so it doesn’t turn in to a tirade.
    1. Doctors tend to want to study, and grant advisory committees tend to want to fund INCREMENTAL research. Anything that is not incremental on something that is being done prominently is considered “high risk”. Good luck with your out of the box thinking. You’ll be labeled a hack and your career will end.
    2. Many doctors who apply for and receive grants from cancer research foundations do so only because it is a good source of funding for their life’s work if they can construe it to in some way have an angle on cancer. A real life example would be studying the immune systems of Zebra fish because: an immunotherapy discovery in the similarly functioning immune systems of Zebra fish might create a model that is applicable to humans. I’ve been told behind closed doors that it’s flat out bullshit, he probably has studied the immune systems of fish his whole career.
    I was flat out asked by the director of a major US cancer research center to just give them money so that the could continue to keep certain researchers on salary.
    Lastly, we want to fund a study into using THC (yes, from Marijuana) as an option for children with cancer because it positively affects three major areas of children with cancer: pain, nausea, and appetite. Right now we pump them full of IV nutrition, anti-nausea meds, and morphine. All of which is very hard on their little bodies. Top cancer research doctors are horrified of the light this kind of research will cast them in so good luck finding someone who will do it, much less sit on your board.
    I could go on forever but I’ll leave you with this. My wife and I lived on PubMed and Wikipedia while our daughter was sick. We wanted to understand everything, and you know what we found? The very first question that every single parent who has a child that is diagnosed with cancer asks: “What can I feed my child?” They have to say “we don’t know.”. Search PubMed, you won’t find a single study related to nutrition and solid tumor cancers in pediatric patients. That’s f*d up, and its because of the bureaucracy and entropy in the research community.
    We’d like to break that mold, but first we’ll need to raise a lot of money.
    Shameless plug http://laylagrace.org

    Reply
  10. A Amar

    “Test the computer program? Non-conclusive (and perhaps not able of being reproduced, as the program should be evolving along with technology).”
    If I understand the problem you’re talking about sufficiently, then in fact, we do have testable “computer programs” in essence; pharma companies have been pursuing this for several years; and I believe it’s a multi-billion dollar business.
    The basic technique, as I understand it (don’t work directly in this field, but have heard from some people who do), is that the “computer program” is a screening process that includes a complex set of tests that identifies how a particular cancer is operating in a specific individual, i.e. what are the specific genes that are causing tumor growth in this individual (when they wouldn’t in another person). Then some computation+database give a list of drug or drug-combination candidates that may be helpful in that case.
    Those individual drugs do require FDA approval, which is a limitation, but many new drugs are getting approval which are specifically useful only for certain rare situations identified via process like the one I describe. Also, in many other cases, an effective therapy can be concocted out of previously-approved chemicals.
    This process itself can be tested much like any other medical technique, e.g. double-blind. It’s true that traditional testing methods would require “code-freezing” the algorithm/process over the course of the trial, but that’s a manageable hurdle.
    Yes, it’s different from regular clinical trials, and there are some business/legal challenges, but it’s simpler in some ways too, since IIRC the FDA does not require regular clinical trials. And the processes are widely patented already.
    So there’s good news here, for the most part.

    Reply
  11. Karmel

    According to my Kindle, I’m 21% of the way through Emperor of Maladies, so I’ve been thinking along very similar lines lately–
    I think one thing, though, that needs to be clarified is that though bioinformaticists and biologists are looking to characterize the individual perturbations that define a myriad of different cancers, the goal is not to develop treatments that will counteract these idiosyncrasies. That is, as you point out, if person A’s thyroid cancer has gene X upregulated, and person B’s thyroid cancer has gene Y upregulated, it would be untenable to try to treat person A by reducing levels of gene X and person B by reducing levels of gene Y, a zillion times over for all the people out there. Instead, the hope is that we can gain a deeper understanding of all these idiosyncrasies, and class them and separate them into distinct functional categories, and then find out what the underlying malignancy is that causes varying but related symptoms across persons A through Z. Then, we could treat the underlying sources of what would be an admittedly large set of problems, but not an infinite set that is completely distinct for each and every individual.
    Incidentally, this problem of generalization over symptoms versus the specificity required for a cure is not unique to cancer; many diseases, such as diabetes, include a variety of underlying causes (i.e., autoimmunity from a number of different genetic sources, metabolic disregulation, cystic fibrosis, monogenic diseases) that can all be grouped according to symptoms and currently available treatments, but that may, in the end, require distinct cures. That said, a cure for one will likely shed light on all– and there is clear economic and political benefit derived from grouping– so the generalizations are not without merit.

    Reply
  12. moioci

    I must agree with John above in one regard: your understanding of scientific method is flawed. We don’t prove hypotheses in the mathematical sense of the word; we test them, and develop evidence that tends to support them or not. Your next statement may often hold true in physics, but rarely if ever in enormously complicated biological systems: “if it works a few times and can be reproduced, it will work always.” Cancer, like medicine and biology, is not like that, true. But cancer and its treatment are products of purely physical processes, which means they are entirely approachable by scientific means of analysis. Finally, if there were a series of techniques to produce therapies specific to a given tumor, that process itself would be testable, just as one algorithm in CS is testable against others.

    Reply
  13. DanielBMarkham

    “…your understanding of scientific method is flawed…”
    I’m going to save to later a debate on my understanding of the scientific method. At that time we can drag out Peirce, and Popper, and Hume, and the rest and have a fine old time. Right now, however, it’s not germane. Apologies for bringing it up.
    The point here, in broad terms, is that there is not a single thing to be tested — not a single disease, a single process, or a single treatment. This creates an epistemological problem — even with the corrections you guys have kindly provided. This isn’t a matter of wanting the precision of physics: this is a matter of not being able to form a single hypothesis, that is, the hypothesis itself is at a meta level. Think of the issues with proving a program correct. Now apply that at the next level of knowledge. We don’t have anything to hold on to. In fact, our desire to have a single thing to test in an epidemiological fashion prevents us from actually fixing the problem, which was the point of the article.
    Thanks for the thoughts, though. Who knew you guys would take me to task using Popper and miss the larger point? Never expected that response. I’ll do the best I can to be more careful in delineating concepts going forward.

    Reply
  14. John

    @DanielBMarkham I wouldn’t want to put anything in personal terms, as moioci does — in fact, I imagine you do understand the scientific method to some extent, but simultaneously hold contradictory and confused opinions about it: exactly as I do, in other words :-)
    The scientific method is in fact central to your one of your two main points (your other point being an economic one). You demonstrate that in your own response to moioci:
    “The point here, in broad terms, is that there is not a single thing to be tested — not a single disease, a single process, or a single treatment. This creates an epistemological problem…”
    That is a point about the scientific method (and the theory of knowledge, if you like), and the way the scientific method is applied in medical trials.
    And again (to risk labouring the point, sorry — but you did object) in your summing up of the blog post itself:
    “I think the way we do science, the way we do medicine, [...snip economic point...] — all of these things don’t match up to the solution we’re working towards in Cancer research.”
    We agree that cancer is objectively and qualitatively different from other diseases, because it involves many different genes and progresses via natural selection. I think you are correct that regulators have a tendency to think in the way you describe in the first quote above, and that that is the wrong way to think. But you have fallen into the same trap.
    Since theories are never tested in isolation, it is a misconception to say that there isn’t a single thing to test. Saying that implies that there sometimes *is* a single thing to test. This may sound like theoretical point-scoring without practical consequence: clearly some tests are more specific than others. But it is exactly this misconception about the scientific method (and other related misconceptions) that leads to the tendency that your blog post is about! The comment that empirical (“black-box”) testing is applicable to complex treatment regimes is half-right: it is possible and useful to perform trials on a complex treatment regime as a whole. Only half-right, because there is no such thing as empirical testing — in reality all observation is “theory-laden” (“white-box”). Whether we can be confident only after testing each gene-specific treatment, or instead confident even without needing that level of testing, depends on our level of understanding of the biology involved. The danger is that our social processes fail to catch up quickly with the fact that as our understanding of biology has improved, so what we need to test changes with it.
    Rather than putting the “let’s test every chemical intervention” tendency down to misconceptions of the scientific method, some people might put it down to simple conservatism. I guess that’s true, but mis-applied convervatism, though a good psychological explanation, can be overcome through rational debate, as long as there is good understanding of scientific method.
    I’m sure you’ll be with me, by the way, in thinking that testing the individual chemical interventions is not without value (taking another software analogy: I’m not saying unit tests are worthless). But there is an opportunity cost to doing those tests that may push up the overall risk to the patient: namely, if over-caution prevents people getting their hands on the treatment, they’ll die for want of the treatment.
    I think you have come to the right conclusion about clinical trials of complex cancer treatment regimes, but for the wrong reasons. The reasons are important because they are the only way to reliably persuade other people that you are correct (and in fact almost the only way to make any progress at all).
    As for the other point you made — perhaps the solution to how to make it economically viable to write computer programs that implement complex medical treatment regimes is the same as in any other part of the software industry: rely on copyright law and prohibit software patents.

    Reply
  15. John

    … I should add that I’m not claiming we are in fact close to being able to let loose with cocktails of drugs, each one of which has not been tested individually. First, I’m fairly clueless about the actual biology. Second, I’m more interested in the fact that there seems to be a more general problem here: The different-drug-regime-for-each-patient tactic is only one of several different ambitious cancer strategies I’ve heard about, and they all seem to cause similar sorts of confusion and rigidity in the social systems we have set up to handle medical treatment.
    Glad to hear that at least some new styles of treatment are actually being put into practice, though!

    Reply
  16. DanielBMarkham

    John,
    Thanks for the long comment. Gave me something to consider.
    I think here’s the problem:
    When I say “This creates an epistemological problem,” I mean in terms of the average regulator or layman, not in terms of a scientist who might understand things at a deeper level. If your tools are crude, your conclusions will be crude as well.
    When I say “I think the way we do science…” by “the way” I mean the way society and governments fund and approve of research, not the Bayesian science that happens everyday somewhere else.
    I think we agree. I am not saying that science is black and white. Far from it. There’s a great argument to be made that a Bayesian model is the only thing we’re left with after several hundred years of discussion. I’m talking about systems of people — governments, regulators, newspaper reporters, and yes, those guys that write the checks for research. You can’t have a comic-book idea of how something works and attempt to govern and control it; at least not without a lot of pain.
    I’m a little lost with our comparisons to computer programs, though. You realize that the role of genes, proteins, DNA, RNA, mRNA, and a lot of other things come into play here, right? A cell may have an “injury” many times over it’s lifetime before it becomes cancerous. Some of those changes may be benign, some may not. It’s the particular combination both of mutations to the DNA and all those other things that cause the cancer. I understand we’re on to biological pathways now, and the other poster had a lot of positive things to say about this research. But my point still holds: we may be beginning to understand the general mechanism of cancer and it may be too complex for our systems of people to deal with.
    Perhaps I’ve missed it, and I should be happy to be right for the wrong reasons. Don’t know. You failed to make a persuasive case, but I’ve been known to be a bit hard-headed. Thank you very much for trying! :)

    Reply
  17. Raphael Lehrer

    Daniel – I found your post to be very impressive. It almost exactly describes what we do at GeneKey. Obviously, we don’t have data on each of the million proteins, but we get as close as we can (hundreds of thousands of data points across multiple technologies – and we are committed to updating as the technology evolves). We’ve found that we’re able to make predictions, and get responses to the drugs in patients. (Not every time, of course.) Of course, many of the pathways we see do not have drugs against them, but we’ve found at least one “broken” pathway that can be addressed by either approved drugs or drugs in clinical trials. You may guess this – but it is not purely algorithm-based at this point. We prefer a mixture of algorithm and human judgment as this allows a faster learning curve – basically, a human can “notice” new things better than a computer, which notices only what you tell it to notice. (By the way, we don’t work on a patient’s case unless their oncologist is on board – and we’ve worked across both community physicians and virtually all of the top US academic medical centers.)
    By the way, let me say that our approach involves rigorous statistics based on one patient. No, it’s not impossible – it relies on triangulation of multiple data points that suggest the same problem. Rather than building them off a significant number of patients looking at a single gene, we build them in a single patient based on multiple semi-redundant genes. For physics/math jocks, you can read about this in much more detail in my pending patent on the topic.
    Right now we’re beginning single-armed studies, but we will be planning randomized double-blind studies in the future.
    I don’t want to prattle on about how it fits in the clinical trial frameworks, or the economics of drugs in this area, or in the regulatory framework. Suffice it to say that things are not nearly as grim as you fear they are (though it took several months of hard thought to sort them out.)
    If this sounds intriguing – particularly if you are interested in how we apply this to today’s patients, but also clinical trial design/economics/regulatory – check us out at http://www.genekey.com or give us a call at (617) 340-6363 and ask for me. The information on the site may be too basic for readers of this thread. I’m happy to discuss more, especially with people who are thinking as deeply about cancer as you are.
    Raphael Lehrer, Ph.D.
    Chief Scientist, GeneKey
    1340 Centre St, Newton, MA 02459
    (617) 340-6363
    rleh...@genekey.com
    http://www.genekey.com

    Reply

Leave a Reply

Your email address will not be published. Required fields are marked *

* Copy this password:

* Type or paste password here:

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>