#Minimizer Fault?
17 messages in this thread
Has anyone tried minimizing a conjugated non-aromatic system? i.e.,
butadiene or an alpha,beta-unsaturated ketone? Using MM2 (the best of the
given force fields for small molecules(?)) the minimized structure gives
equal bond lengths for both the C1-C2 (or C3-C4) double bond and the C2-C3
single bond. What's going on here? If you submit the structure for
minimization in CNDO/MNDO then the minimized structure seems to be correct
with the calculated single bond length very close to the experimental
value. Haven't yet gotten a reply on this from Autocad (and/or Hypercube –
can't figure out who's going to be responsible for supporting such
questions). Any structure fitting this general pattern seems to suffer the
same problem, i.e., diacetylene also minimizes with all bond lengths equal.
Am I missing something here?
Stan,
The MM+ force field, which as you say is the one to use for most
non-biomolecules, is closely the MM2(1977) force field of N. L. Allinger.
Probably the biggest single weakness of this force field (as well as all
others in which bond characteristics are assigned solely on the basis of
the atom types involved in that bond, which includes many different force
fields), is just the one you point out. Bond lengths are assigned on the
basis of atom types, and consequently you get an inaccurate treatment of
conjugated (non-aromatic) bonds.
There are some ways around this in the development of force fields, such as
adding bond order to the information used to define bond characteristics,
or add some non-molecular-mechanics feature to the algorithm, such as a
quantum-mechanical description of the pi electrons. We are currently
looking at alternative ways of improving this force field compatible with
our intent of using only published, public-domain science for reasons of
reproducibility.
Depending on your application, I can suggest a couple of things: 1). The
model builder gives pretty good structures for these systems. If all you
are interested in is structure (not energetics) then selecting and buildig
the conjugated part would give you reasonable bond lengths. 2). The
semi-empirical methods do, as you say, give good structures and energetics
for conjugated systems, as well as most others. They take longer, of
course, but for smallish systems may provide what you want.
The stereochemistry of conjugated systems is pretty well-defined, so that
inaccuracies in these bond lengths are very unlikely to have global effects
on the structure of your molecule, of course.
On your other point, the support mechanism for HyperChem is the same as for
most other Autodesk products: the dealers (value-added resellers) who sell
the product have responsibility for support. The dealers in turn get
support from Hypercube.
Hope this helps,
Tom Slee, Hypercube.
Thanks for your quick reply. Actually it was fairly obvious that the
minimizer was relying upon the atom type definition I was really trying to
make sure that there was no workaround that I was missing. I think that
MM2(85) fixed this problem. Is there a time schedule for updating the
force field in Hyperchem? We are primarily attempting to understand the
placement of your software against its competitors. Even at the academic
price (which we don't qualify for) there are a number of PC packages
available which include more sophisticated versions of MM2. I like your
reliance upon publicly available techniques but it may also serve to
differentiate your software from some others. You compound the problem
with the two tier pricing scheme. At the Industrial price you really start
to compete against some state-of-the-art stuff (albeit usually with
proprietary force fields and maybe running on different platforms (RISC
6000 or MAC) but the hardware usually isn't a problem for us. As far as
the support question is concerned, we're not waiting (about 2 weeks) for an
answer back through the distributor channel (did get several calls from
Autocad sales people, but no technical help). During this our AUTOCAD
regional sales person called, asked us to work directly with AUTOCAD as we
were a "preferred account" and promised a call back within an hour. Still
no reply. I'm not really complaining just relating the events. I do
appreciate your prompt reply and useful information. Like I said, we're
really just trying to decide whether we should promote it for internal
adoption or not.
Hi Stan, I'd like your thoughts on the following.
I am an industrial chemist who would like to learn a little something about
modelling and would also like to use a good program to do some actual work,
starting with small systems and working up. However Hyperchem is way too
expensive for a private individual and as an industrial chemist it is
almost impossible to persuade the very cost conscious people "above" me to
get this product. I could afford it if I could buy at the academic price
but not at the industrial.
In my opinion:
1.) Real value has yet to come the desk top as far as molecular modelling
is concerned. Most packages out there are (for the PC) are toys, I mean ,
really… runs in 640K (MOBY) , one force field, limited number of atoms,
BAD (DOS !!!) graphics , or …. (Chem-X form Chem Design)… rediculously
expensive (4 x 4000 per year),
cant use extended memory (PC Model… unless something has changed big
time)….
2.) Autodesk, I believe has severly misunderstood the market for this
package. I have some indirect evidence for this in that discussion with
certain people from that company I have been informed that they have made
poor headway in industrial sales… at 5X the academic price , no wonder.
They are too elitist and far too insular… too influenced by their
succcess with autocad….
3.) Some numbers: at $ 100 evryone would buy… thats coffee money.
$300.. almost everyone would buy even as individuals, $500 … everyone
who understands what their getting will buy. $750.. some will try to shop
around but if the product is really superior those in the know and who are
serious will still buy , but delay their purchase.. 70-80 % of indiv will
still purchase from own budget.
NOW…. 900 here hard resistance and non linear sets in and individual
purchasers will drop out to the 50% level and some will sacrifice features
for a 500 package instead.
1000– this is getting really expensive for a PC package serious will go
for it but delay purchase.
Edmund,
Can I chip in on this one? You suggest that Autodesk is pricing HyperChem
too high, and that this reflects elitism. I really don't think this is so.
There seems to be a feeling that if a program runs on a PC it should be
priced much cheaper than if it runs on a workstation, because it must
somehow be a toy. In the case of HyperChem I really don't think this is
so. The ability to do molecular dynamics and quantum mechanics on a PC in
a big way is, I think, new with us. The ability to use selections to do
calculations on part of a system, or to incorporate restraints into
optimizations and dynamics, makes it more sophisticated than other PC
products, and the scripting (and hence DDE) capabilities I think are also a
step up on other PC products.
Current PC's (486-50's, say, or even 486-33's) provide computational speed
that is up to that of workstations a few years ago, and with the proper
memory management large systems can be tackled. I think HyperChem is
priced much cheaper than other products with the kinds of facilities we
offer; hardly an unbiased observer, of course, but I do think this is so.
The fact that these other programs run on workstations does not make it
natural that they be more expensive.
Ah well, that's all my opinion, of course. I guess I do think that real
value can come from desktop modeling with HyperChem and that makes it worth
the money. Whatever, I hope you make the best decision and can, if it is
HyperChem, persuade those cost-conscious ones of it.
Best wishes,
Tom Slee, Hypercube
>> I have been informed that they have made poor headway in industrial
sales… at 5X the academic price , no wonder
Ed,
Almost any software will be more slowly accepted into industrial circles,
who tend to make group decisions and have standards to recommend or
enforce. Academics will buy quickly and individually. Industrial sales
take more time — but no academic institution comes up with 100-unit
orders, either.
I'm not with Autodesk (we're a 3rd party developer), and I see some areas
where their marketing strategy could be adjusted. But there are also some
outstanding areas in their marketing plan as well. It's too early, IMO, to
tell if their industrial sales will be okay or not. Unlike AutoCAD, which
sold into a market (architects, etc.) where there are many small firms,
chemistry is a small number of large firms, which makes it hit or miss.
I don't see their competition as being PC products you mentioned such as
Chemical Design's ChemX, or especially PC Model (or things like Tripos'
Alchemy). I think that they are competing against Tripos' larger stuff,
Chemical Design's larger stuff, Biosym, etc. The issue becomes one of
whether you outfit a few heavy workstations for shared use, or put
HyperChem on everybody's desk. Either way, the solution runs well into six
figures.
HyperChem's DDE and scripting capabilities make it easy for it to snap into
a larger solution. Tie it into Excel using DDE, and you can quickly start
creating "roll your own" QSAR applications. That should threaten companies
like BioCad.
Then, when you consider that Autodesk announced a version for SGI, you get
a step up in performance and compatibility between the PC and SGI
platforms.
>> 1000– this is getting really expensive for a PC package
Me again Richard, I agree with your comments, overall. But want to point
out that in many instances (particularly the "small number of large firms")
modeling solutions are usually available in some form (on workstations,
mainframes, home-grown software etc.) i.e., the initial relatively large
investment has already been made and amortized. The PC packages are
getting better, and I think that Hyperchem has upped the ante considerably,
and I think that you're right it is going to take some time for us
industrial types to decide where and if we're going to place Hyperchem
somewhere in the line-up.
Stan,
It will be interesting to see if a common model or models develops for
positioning of HyperChem within companies with existing expensive modeling
packages, or whether the models are very sensitive to the particular
company. During our testing program I saw a couple of places using a
two-tire system of sorts. They use their expensive package on the
workstation, and got several copies of HyperChem to run on PC's. They move
data back and forth using the PDB file format, and get to relieve the
congestion at the workstation in an efficient manner. I have seen the same
model talked about at a couple of places for when the SGI version of
HyperChem comes out (soon!); the superior graphics and processing speed of
the workstation can be saved for when they are really needed while the PC
version can be used for a large percentage of the work.
I will be interested to see what models do develop elsewhere.
Tom Slee, Hypercube
I think such a two-tier system is where we'll end up with most users not
bothering with large packages. In this regard I think the visualization
tools become really important, In fact they may overwhelm the modeling
aspect of the software. Electronic Drieding models (much more flexible and
capable of being queried as to approach distances, etc., would appeal to a
lot of traditional chemist types. I think the Builder aspects of Hyperchem
are pretty interesting in this regard.
We're beginning to promulgate a number of RISC 6000 machines on our
network. Any plans for a RISC version of Hyperchem? Is the SGI version
network aware? Is it practical to run it remotely (say, from a PC accessing
the workstation over ethernet)?
I enjoyed reading your comments. As a value added HypeChem dealer we're
certainly having some success. But we also find that the challenge for
HyperChem is finding its position in the market, not something that happens
overnight. Many industrial institutions, especially pharmaceutical
companies, have big-time modeling investments. However, serious modeling
has often been left to the computational chemist. We find this is changing
as more chemists/biochemists/immunochemists, etc., are being exposed to
computational tools like MOPAC, Sybil, Chem-X, etc. Many chemists,
though, have been turned off by ealier PC packages that have severe
limitations and utility. HyperChem certainly is the 1st of a new
generation of PC/desktop systems and has to overcome the poor record of
other PC systems and find its place relative to mini- or mainframe systems.
We also are finding two kind of industrial customers: those who have not
had good experience with prior PC systems and therefor feel HyperChem is
too expensive, and those who have "big system" experience and can't
believe that HyperChem can possibly work at ONLY $3500. We are undeterred
by either because we think HyperChem, for what it is, is a good product
with top notch documentation.
I think our own interest is certainly driven by bench chemists, and you may
see my other comments to Richard Hong and Tom Slee about the benefits I
think are to be gained by moving modeling packages to the bench chemist.
We've also seen the PC packages sneered at by the big-package user, and
I've personally struggled with poor PC packages (poorly written, and poorly
executed science) – I think I've got every PC package available. Hyperchem
does push the envelope with regard to PC packages, no question about it.
One of the hurdles we'll have to deal with is gaining Hyperchem some
credibility against the large packages (the issue you raise in your note).
The industrial price is large enough so that it gets questioned (the list
price puts it above our ability to casually order the thing. It becomes a
"capital" purchase that entails an additional layer of bureaucracy), we now
have 2 copies in-house and expect to give it a fairly exhaustive
evaluation.
>> in many instances (particularly the "small number of large firms")
modeling solutions are usually available in some form
Stan,
I guess my question is whether the chemists in these firms feel that they
have decent control over the use of those larger modeling systems, or
whether they have to cede control to computational chemistry specialists.
It is my understanding that in some firms they have computational chemistry
groups that function as service groups (just like an analytical service
lab) for other chemists.
If the chemists feel that they do not have cost-effective access (what are
in-house time charges for the larger systems?), they might buy HyperChem to
save money. It wouldn't be the first time that in-house cost structures
compelled people to spend cash money in order to save on internal budget
money <g>.
Personally, I could see HyperChem taking advantage of any possible friction
between the bench chemist and the computational specialists the same way
that Macintoshes played on friction between end-users and MIS people.
Rich
I think you are very right, I find myself making these same arguments.
Rather than focus on "friction" and "control" issues however (try to
discuss these issues in strategy meetings with both ends of the argument
present!) I tend to emphasize that there are substantial benefits to be had
by moving the modeling tools to the lowest common denominator in order to
place the tools in the hands of the bench chemist. This is not without
problems, a lot of people don't understand the nature of the minimization
process and searching for global minima (just one example of the type of
complexity that can be involved) is far from trivial. Too many people
still believe "if it's the result of a calculation then it must be right".
Really a training issue.
I view modeling as just another tool in the arsenal. Don't expect it to
answer all your questions, but if it can prioritize or eliminate some
experimentation for you then it's all worth it. Even if the packages are
only used as visualization tools (electronic Dreiding models, the build
module in Hyperchem is really interesting in this regard) if they work for
you, then they are justified.
There's another fallout, I think that the demand upon the modeling service
group will increase as the tools promulgate throughout the organization
(can only help their job security) and they'll be asked more intelligent
and more appropriate questions.
Can't overemphasize the size of the training problem though.
>> Too many people still believe "if it's the result of a calculation then
it must be right"
Stan,
You are absolutely right. I've been doing some playing with HyperChem vs.
Alchemy III, and numbers on some calculations just don't match up. It
seems that Alchemy is using a lot of shortcuts in order to produce poorer
results faster.
I think that Alchemy has some cute renderings however, especially simple
ones (ball and stick, cylinders) while HyperChem seems to have concentrated
on the higher-end renderings. I kind of like the Alchemy "joystick" grid
for moving the molecule, though I have trouble controlling the movement
sensitively.
I agree 1000% about the training issue. Inadvertently, HyperChem is going
to allow a lot of bad modeling to be done! A little knowledge (or in many
cases, no knowledge at all) will no doubt be dangerous.
Rich
(cont.)
What's funny to me is that I believe that if you took HyperChem, bought a
Dell 486, put the whole thing in a black box (dedicating the Dell to *only*
run HyperChem) and sold the whole thing for $7,500 there'd be much less
questioning of the price tag.
Because it runs on a PC, it's supposed to be cheap. The flip side is that
to get similar functionality, you'd have to buy a $10K workstation and put
another $10K software package on it. Instead, you can add a $3500 addition
to your 386/486 PC, and get just about the same thing done.
AutoCad managed to dominate the CAD market despite a flurry of low-end PC
CAD packages. Basically, AutoCad was/is a better product, and was aimed at
the people who *need* it.
I don't know how well HyperChem will sell to people who have never used
modelers before. I suspect that you and I would agree that first-time
users will not be a big market for this product. Personally, I think that
its best market will be at companies who have 60 chemists sharing 2
dedicated UNIX workstations for modeling. Chemists there might get tired
of waiting to access a machine, and purchase HyperChem.
But my assumption is that their prime market is chemists who have tried
(or wanted) expensive modeling software running on big boxes and can be
convinced that if they move down to HyperChem, they lose little.
Rich
I agree with a lot of what you say, which is why I'm really interested in
what value is added by Hyperchem (particularly since the industrial price
really pits the software against some software of good functionality
(albeit they tend to run on platforms other than PC (RISC, primarily)). PC
modelling software seems to fall within two broad categories:
1. Written by Chemist (physicist or whatever). Usually gives good
scientific results but few of these packages (PC MODEL is a good example)
are written by software experts so performance is relatively poor
(particularly memory usage and graphics manipulation).
2. Written by software mavens who don't seem to do a good job of the
science but produce packages that do a better job living up to the current
standards of PC software.
By the way, ALCHEMY III software is an interesting middle ground. Good
use of PC and nice programming (particulary graphics rotation, etc.,
especially the DOS version) but the program hasn't got enough real modeling
capability to be of much serious use (e.g., no ability to do even the most
rudimentary conformational searching, no torsion driver for example).
Hyperchem holds promise overall, I think. I'm hoping it will mature
relatively rapidly and expect that support bugs will get worked out soon,
but I still need to get its position (capabilities and price) relative to
other packages straight in my mind.
Price is always a relative thing, its difficult to demonstrate the value
derived from any software and modeling always seems to bring along a lot of
baggage of its own. Let me know what you think.
By the way, PC Model is now available in a 386/387 version that I believe
makes better use of PC resources.