Gina Miller writes:
> "Chemists have no good algorithms for manually enumerating the molecules
> that can satisfy a set of constraints in 3D," he said, in his March 5th
'Manually'? Isn't there such a thing a combinatorial explosion? Are we supposed to think up billions of combinations and screen them by building molecular models and docking them by hand?
> presentation at the NAS Computational Molecular Nanotechnology Workshop.
> Wipke believes that "computers, when properly programmed, [can] be more
> creative than chemists in the task of inventing new chemical structures."
Wow. What a profound statement.
> Called INVENTON, the program automatically constructs chemical frameworks
> [...]
> scoring functions selected by the chemist, Wipke explained.
All this is not exactly new. Look at Gasteiger's work, for instance. WODCA even gives you synthesis routes.
> Productive Approach To Design
> "A computer program can invent molecules that have never been conceived of
> and do not exist in any database anywhere," he commented. In its six years
> of existence, INVENTON has designed structures in the domains of
> nanotechnology (self-assembling molecules in nanoscale systems) and
> pharmaceutical drugs (such as mimics of known drugs for cancer and enzyme
> inhibitors for AIDS).
> "Designing a system like this is a complex task -- one that requires really
> logical thinking. AI methods are already being used by INVENTON to
> prioritize tasks, and heuristic search is used to reduce the number of
> `relevant' constraints." In addition, bringing "engineering principles
> across boundaries to assist chemical design" has been useful, Wipke noted.
I don't see GA methods mentioned here. I don't have much faith in AI methods in that context.
> Automated molecular design will benefit from greater CPU speed, more
The thing won't run parallely in a cluster?
> knowledge and understanding, and larger organized collections of chemical
> information readable by computer, he said, adding that "having more complex
> problems will accelerate our abilities to problem solve."
They don't tell whether the system does learn progressively or has to start from scratch each time.
> Collaboration with NAS
> Collaborative work between Wipke's group at UCSC and Al Globus (NAS
> applications and tools group) has included modifications to the search
> strategy and solving simple initial problems. Current work is focused on
> developing fragment libraries designed for nanotechnology and then applying
Now that particular focus is indeed interesting.
> them to problems requiring very stiff molecules. A more difficult future
> problem is taking advantage of the high degree of symmetry common in
> nanotechnology problems and spanning space with molecular fragments that can
'High degree of symmetry common in nanotechnology problems'? Wow. Nanotechnology must be way further than I thought. Or maybe they have got a really good crystal ball.
> be repeated an indefinite number of times.
>
> http://science.nas.nasa.gov/Pubs/NASnews/96/08/invention.html
Thanks for the item. Why don't you post such stuff to transhumantech@excelsior.org ? We have need of such content.