Dear Joe, dear FISErs,

An organic chemist is able to predict a number of properties from the
structural formula, including much about reactivity of the compound.
But as you know, doing that properly is extremely difficult in a
number of cases, because the rules governing reactivity are much more
complicated that the ones which are taught at Universities, and the
number of rules expands rapidly each year. In fact, an experienced
Organic Chemist has in his head a so extraordinary rich collection of
rules and a so enormous knowledge that even many chemists which are
not Organicians cannot imagine the extent of this knowledge.
It is clear that the "doing chemistry" process derives from these
rules (these rules are chemical information), not only from the
Since the 70's, some cheminformaticians tried to store that in
databases: reactions databases plus databases of reactivity rules for
computer sssisted synthesis or retrosynthesis, etc., then built
programmes intended to output proposals supposed to help the chemist.
As far as I know, the brain of the Organician is still by far much
more efficient than the best softwares which were produced.
So, I may tell that the information available in the brain of the
Organician is extremely difficult to store on computer, and it is even
very difficult to teach it, apart the very beginning.
There are examples other than reactivity. A huge of QSAR studies were
done in order to predict various physico-chemical properties of simple
chemical compounds, e.g., predicting from the structural formulas the
boiling temperatures of monofunctional compounds such as alcohols,
cetones, etc. at 20 C under 1 atm. But even in these apparently simple
cases, the chemical information we need to do that with an acceptable
accuracy is difficult to extract: the conclusions of such QSAR studies
cannot be applied to any alcohol or cetone (still assumed to be
monofunctional compounds), and it is even difficult to know the extent
of validity of the published empirical rules, concretely often
summarized by some regression coefficients.
The example of spectroscopic databases is also of interest. How
simulate spectras (infrared, NMR, mass spectras, etc.) of chemical
compounds ? Starting from the structural formula, it is really hard to
simulate, e.g. a low resolution mass spectra. Most time, it was
attempted to extract rules from spectroscopic databases, then try to
predict the spectra of a compound absent from the database, or
conversely, retrieving the structural formula of a compound from its
spectra(s). Many such softwares were developped since the 70's (one of
the oldest ones is STIRS), but really the chemical information needed
to do that properly is very difficult to extract.
To conclude, I retain your example of crystallization: for sure when
we will able to retrieve from the structural formula H-O-H that water
under 1 atm should crystallize at 0 C, then for sure we will be ready
to predict more about crystallization of chemicals.

Best regards,


2011/9/17 <>:
> Dear Michel and FIS Colleagues,
> This will be an interesting discussion, since the core nature and role of
> information will be involved. Here is just one first point: to me, as a
> chemist, chemical information is only secondarily an "object" capable of
> being formalized, archived, etc. A formula has meaning for me in terms
> of the potential reactions the molecule to which it refers can undergo, what
> it looked like when crystallized for the first time and so on.
> Cheminformatics seems not to deal with such aspects of chemical information
> as part of a process of "doing chemistry". Can this be captured by  another 
> system?
> Best wishes,
> Joseph

fis mailing list

Reply via email to