Search Mailing List Archives

Limit search to: Subject & Body Subject Author
Sort by: Reverse Sort
Limit to: All This Week Last Week This Month Last Month
Select Date Range     through    

necessary & sufficient

SLetovsky at SLetovsky at
Tue Feb 17 12:22:38 PST 2004

In a message dated 2/16/2004 4:42:24 PM Eastern Standard Time, 
jblake at writes:
Hi Stan,

I think David illustrates well that biological knowledge is the 
accumulation of experimental information that allows a 'judgement' ;  there 
is not a 'necessary and sufficient' component of a GO term association that 
would be easy to define, I think. The evidence codes provide a method of 
'summary statement' of the evidence used in standardized way that allows 
users a 'confidence' assessment in a way. The citation provides the 
researcher access to the information about the experimental procedures 
used, etc.

What I was after was a general recipe for testing predicted GO assignments. 
So the
evidence for known GO assignments doesn't really help. David's criteria are 
on target, but they are quite complex, and fundamentally genetic. They 
suggest (to me)
the following N&S conditions:

    gene G gets biological process term T iff
    there exists a homo/heterozygous loss/gain-of-function mutant for G, 
                      a genetic background
               and an environment 
    that together show a perturbed T which is not attributable to some other 
process U 
    which is causally upstream from T but not "part-of" T,
    and which is not evident in the same genetic background and environment 
    the wild-type G.

(I am trying to make the definition handle lethals and pathway redundancy 
correctly. Both cause problems -- lethals because they affect everything, yet we 
don't say they are involved in all processes, and redundant pathway steps 
because they may not have a pheno in wild type, yet we still may say they are 
involved in some processes. So the "which is not attributable to" is for lethals, 
and the "in some genetic background" is for redundants.)

To use this definition for experimental design would seem to require:

    1. ability to generate KO / OEX constructs for any gene of interest
    2. screens/phenotypes for any process T, sensitive to "direct effects" on 
T only
    3. a potentially unlimited set of background genotypes (to detect 
interactions) and 
      environmental conditions

1&2 is not inconceivable given high-throughput screens that have been 
3 is much more problematic. Part of the beauty of the "module networks" paper
was that (2) was always microarrays, and (3) was specified by the
automatically generated hypotheses. 

David's criteria make sense, but I am somewhat surprised at how "genetic" 
they are --
not that it is surprising that David would propose genetic criteria, but 
rather that
purely phenotypic criteria are used to associate a gene with a biological 
I always thought part of the goal of GO was to move away from describing
things in terms of mutant phenos to associating genes with normal biological
processes, but maybe that's just a terminological shift. Does this mean that 
there is (in a given species) always a clear mapping between phenos and BPs?

Cheers, -Stan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <>

More information about the go-friends mailing list