[Comp-neuro] Re: Reproducability, Funding, and a note on hippocampus

A. David Redish adr at adrlab.ahc.umn.edu
Tue Aug 19 14:24:12 CEST 2008


1. Reproducability.

I want to add one quick point extending some points made in the recent
discussion.  Computational models are not the only place where
neuroscience has problems of reproducability, in fact, I would argue
that it is one of the places where reproducability is actually the
best in neuroscience.  (You can always send completed code, I can run
it, and then I can deconstruct it if I want.  It's kind of hard to
send me your behaving/recorded rat...)  In most of the rest of
the field of neuroscience, there are many many subtle (and often not
understood) critical factors.  (I know of one case where the
anesthesia used in the implanting of chronic electrodes in whisker
cortex made a difference days or weeks later when the labs in question
recorded neurons from that cortex.)  Describing all of the training
details (did the animals have toys in their cages or not) is almost
never complete, even in the best of papers.  

Also, one of the areas computational neuroscience has been pushing
experimental neuroscience recently is in the complexity of analyses.
These days, there are results that are only visible after significant
(and complex) post-processing of data.  Particularly in the fields
where I work (fast dynamics of neural ensembles), there are important
results that are simply not visible at the single-cell level.  I
suggest that we need a method for sharing analytical algorithms as
well as computational models.  It would also be good to find
quantitative and accepted ways to validate and justify algorithms (and
to identify their limits).

2. In terms of venture funding.  

I would agree with the comments about funding and "groundbreaking"
research, but I would point out a few things.  First, neither Planck
nor Einsten were actually funded until after they won their Nobel
prizes.  However, they were well known within the field as people to
keep one's eye on.  (Remember, the correct explanation of Brownian
motion was Einstein's thesis [I think].)  But it is very true that the
current funding mechanism actively discourages transitions between
highly specialized fields.  (For example, I can say from personal
experience that it is nearly impossible to shift one's research from
hippocampus to striatum, even within the same species, using very
similar techniques.)  I think this is more due to the high bar of
preliminary data more than anything else.  (At one point, I
experienced the wonderful catch-22 of "publish a paper that is
essentially your first aim and we will fund you".  This is, of course,
hard to do when one is out of money.)

PS. Referencing and earlier topic: Hippocampus

Neil Burgess is absolute right that neither phase precession nor grid
cells were predicted.  I would note that if anyone had suggested such
existed, they would have been laughed at.  Certainly they would never
have been funded to go look for them!  I think this is an important
statement about the importance of exploration over pure
hypothesis-testing, but that's another issue.

But hippocampus is a place where there has been a tremendous success
in the interaction of theory, modeling, and experiment.  There are a
number of major cases where a theoretical prediction was directly
tested and confirmed (one example is place field expansion).  It has
also been a place where theories and models have been rejectable (for
example, the location of the origin of phase precession).  This
interaction is due to the tight integration between experimental,
theoretical, and modeling work in the field.

adr
- -----------------------------------------------------
A. David Redish         redish at umn.edu
Associate Professor     http://umn.edu/~redish/ 
Department of Neuroscience, University of Minnesota
- -----------------------------------------------------



More information about the Comp-neuro mailing list