[Comp-neuro] useful models and the scientific method

james bower bower at uthscsa.edu
Mon Aug 11 18:19:39 CEST 2008


In the early days of physics becoming what Kuhn* called a paradigmatic  
science ( http://en.wikipedia.org/wiki/The_Structure_of_Scientific_Revolutions) 
, or in other words, a science where a theoretical substrate exists  
that can serve as the basis to knit experimental and theoretical work  
together (or even force them together I would suggest), most  
practitioners were both experimental and theoretical.  I would claim  
that the segmentation between experimentalists and theorists that we  
tolerate today in biology is another of the unfortunate consequences  
of the success of modern physics - and the effort to apply the  
pedagogy of modern physics to Biology, which is still fundamentally a  
pre-paradigmatic science (by Kuhn's definition).  I believe that the  
lack of an underlying theoretical structure in biology makes the  
segmentation into theoretical and experimental tracks very difficult.

Following up on Todd Troyer's remarks (Todd incidentally being a  
theorist who decided it was essential to also be an experimentalist),   
I agree that many if not most scientists of all types don't really  
have much of an understanding about how science really works - in this  
case codified in something we teach to 5th graders as "The Scientific  
Method" and actually seem to believe ourselves.  And also agree, you  
have to look no further than expectations at NIH or other funding  
agencies for "hypothesis-driven research" to see that they also don't  
understand how science really works.  The "hypothesis" that are being  
tested are most often not much more than the 'tyrannical ideas"  that  
abound in neuroscience (cortical columns, Marr/Albus cerebellar  
learning, the significance of synchrony in neuronal coding, etc etc).   
Unfortunately, few and far between are PhD requirements or even  
courses in epistemology or history of science.  As Kuhn and other  
philosophers of science have pointed out, bending history seems to be  
an essential aspect of how science works, whether paradigmatic or not.

For several years I have taught a course to graduate students called  
"The History of Your Science".  In the course, students pick a classic  
paper in their field, then we read the paper together (a significant  
accomplishment in and of itself -- how many of you have, for example,  
actually read Donald Hebb's book (which could never be published  
today), or the classic papers referenced in the first or last  
paragraphs of almost all scientific papers?).  Then in the second  
section of the course, we read several modern papers referencing the  
classic work.  I actually start the course off myself reading  
Mountcastle's original cortical column papers.  Needless to say, the  
scientific process looks much different than most students expect.

Finally, I appreciate your continued indulgence - but these themes are  
also considered in a review I wrote last year for "The American  
Scientist" of a book titled "23 Problems in Systems Neuroscience"  
which resulted from a meeting organized in 2000 whose purpose was to  
generate a "roadmap" for neuroscience comparable to Hilbert's effort  
in mathematics 100 years earlier.

http://www.americanscientist.org/bookshelf/pub/math-envy

In my opinion, the articles in that book make it very clear how far we  
have to go in Computational Neuroscience.


Jim Bower




On Aug 7, 2008, at 3:12 AM, Neil Burgess wrote:

> Re: the discussion of 'realistic' and 'useful' models.
>
> In practice a useful model is one that makes predictions which
> are novel and feasible enough to convince an experimenter
> to actually test them. This is actually quite rare, and
> may not have a simple dependence on either the level of
> biophysical detail or the mathematical elegance of the model.
> (The same is true, in reverse, of useful experiments:)
>
> Best wishes,
>
> Neil
>
> Neil Burgess,
> ICN, UCL.
>> -----Original Message-----
>> From: comp-neuro-bounces at neuroinf.org [mailto:comp-neuro-
>> bounces at neuroinf.org] On Behalf Of jim bower
>> Sent: 01 August 2008 13:45
>> To: bard at math.pitt.edu
>> Cc: comp-neuro at neuroinf.org
>> Subject: Re: [Comp-neuro] From Socrates to Ptolemy
>>
>> Ah Bard,  here I was happily headed back to the ranch (literally)  
>> willing
>> to let the conversation die back ...  But ...
>>
>> Obviously, a useful model is a useful model regardless. And good  
>> science
>> is good science regardless.  however, it is clear from the history of
>> science that different approaches come with different costs and  
>> benefits,
>> and that different approaches are more or less useful depending on  
>> the
>> state of the field. I believe that neuroscience today is more like  
>> physics
>> in the 16th century than like physics in the 21st, and needs to go  
>> through
>> a similar process of finding the appropriate methods for the  
>> appropriate
>> questions. As then, I think that accomplishing those objectives will
>> require that we stay very close to physical reality (as Newton did in
>> using the moon's movement around the earth to both invent (or  
>> borrow) the
>> calculous and discover the inverse square relationship in  
>> gravitational
>> attraction).
>>
>> But, of course, then Newton and his predicessors especially, were  
>> stacked
>> up against the methods, sucess, and vested interests of the catholic
>> church. In some ways I feel we in computational neuroscience are  
>> similarly
>> stacked up against the high priests of science, the physicists, and  
>> their
>> tried and true methods and no doubt valuable set of lessons  
>> learned. But,
>> biology is different and the difference and the conflict is perhaps  
>> best
>> indicated in the difference between abstracted models and "realistic"
>> models.
>>
>> First, I would define realistic models not only as those that  
>> include as
>> much of the actual structure as possible, but also and perhaps most
>> importantly, models that are "idea" nuetral in their construction. Of
>> course I know that in the absolute there is no such thing, but  
>> there is a
>> fundamental difference, for example, in taking 4 years to get a  
>> Purkinje
>> cell model to respond as a real Purkinje cell to current injection,  
>> than
>> starting by assuming Purkinje cells are Marr/Albus learning nodes and
>> proceeeding to build the model accordingly.
>>
>> Second with respect to the 4 years to build the initial model (and  
>> up to
>> now the almost 15 years and counting to understand it), for physics  
>> and
>> abstract models, the larger the number of parameters, in principle,  
>> the
>> easier it is to get the model to do what you want (many famous  
>> quotes on
>> this). In contrast, in realistic models, the larger the number of
>> parameters, the harder it is to get what you want. Further, whether  
>> one
>> knows the exact value of the Kchannel conductances or not, one  
>> knows for
>> sure the likely range, and therefore both GENESIS and NEURON can  
>> provide
>> constraints and in effect alerts to parameters widely out of range.
>>
>> But most probably important for the power of realistic models, they  
>> almost
>> immediately allow one to quantify ones ignorence by indicating  
>> which of
>> the parameters require more data. Being realistic, the requested  
>> data is
>> already in a form that, in principle, can be directly addressed
>> experimentally (I.e. What is the spatial relationship between  
>> excitation
>> and inhibition on the small dendrites of the Purkinje cell.).  That  
>> said
>> one of the tricks in realistic modeling is often using the model to  
>> figure
>> out how to get at a critical parameter indirectly, even if there is
>> currently no experimental technique to get at it directly.
>>
>> Thus, as in physics then and now, the real value of all models  
>> should be
>> to organize experimental science and force experimentalists (and  
>> modelers)
>> to develop new techniques. The more realistic the model, the more
>> immediate the translation to reality.
>>
>> I will say again, however, if the assumptions of function are already
>> built into the model, this is much less likely to happen.
>>
>> So models are a device to get from here to there. Realistic models  
>> make
>> the effort to have this path directed by the structure itself.  
>> Abstract
>> models have often only begat new abstract models (almost all, as  
>> iin the
>> historical case of Ptolemy, more complex than the previous). I hope  
>> we can
>> avoid needing to reach the point as happened in the early history of
>> modern physics, that the shift to realistic models was driven by  
>> the fact
>> that the abstract model had become more complex than the realistic
>> alternative.
>>
>> Finally, again, the purpose of modeling should not primarily be to
>> demonstrate what we know or believe, but to reveal our ignorence  
>> and then
>> direct our progress towards reducing that ignorence . Realistic  
>> models in
>> our hands have always helped us to understand that we know less  
>> than we
>> even thought we did when we started building the model. .
>>
>> Jim
>>
>>
>>
>>
>>
>>
>>
>> I promote and defend realistic modeling. think appropriate methods  
>> it is
>> an interesting and important question,
>> ------Original Message------
>> From: G. Bard Ermentrout
>> To: James Bower
>> Cc: comp-neuro at neuroinf.org
>> ReplyTo: bard at math.pitt.edu
>> Sent: Aug 1, 2008 6:28 AM
>> Subject: Re: [Comp-neuro] From Socrates to Ptolemy
>>
>> We've established that there is no "noise" in the nervous system.  
>> Now lets
>> take on the shibboleth of "realistic" models. So, I will ask you all
>> why a model with 10000 compartments with dozens of active channels,  
>> none
>> of which has been measured (or probably can be with current  
>> techniques) is
>> more realistic than an abstracter model about which one can prove  
>> or argue
>> with some rigor is capable of explaining the underlying phenomena.  
>> I think
>> one can easily go to far in simplifying, but one can also err in the
>> opposite direction.
>>
>> Bard
>>
>>
>>
>> Sent via BlackBerry by AT&T
>
> _______________________________________________
> Comp-neuro mailing list
> Comp-neuro at neuroinf.org
> http://www.neuroinf.org/mailman/listinfo/comp-neuro




==================================

Dr. James M. Bower Ph.D.

Professor of Computational Neuroscience

Research Imaging Center
University of Texas Health Science Center -
-  San Antonio
8403 Floyd Curl Drive
San Antonio Texas  78284-6240

Main Number:  210- 567-8100
Fax: 210 567-8152
Mobile:  210-382-0553

CONFIDENTIAL NOTICE:
The contents of this email and any attachments to it may be privileged  
or
contain privileged and confidential information. This information is  
only
for the viewing or use of the intended recipient. If you have received  
this
e-mail in error or are not the intended recipient, you are hereby  
notified
that any disclosure, copying, distribution or use of, or the taking of  
any
action in reliance upon, any of the information contained in this e- 
mail, or
any of the attachments to this e-mail, is strictly prohibited and that  
this
e-mail and all of the attachments to this e-mail, if any, must be
immediately returned to the sender or destroyed and, in either case,  
this
e-mail and all attachments to this e-mail must be immediately deleted  
from
your computer without making any copies hereof and any and all hard  
copies
made must be destroyed. If you have received this e-mail in error,  
please
notify the sender by e-mail immediately.










More information about the Comp-neuro mailing list