For example, using Abalone.

But why atom by atom? This is usually done by residual.

http://www.biomolecular-modeling.com/Abalone/index.html

I think Kalzium is the best package in order to build a molecular structure atom by atom. For the case of big molecules, optimization process goes a little bit slow depending on your computer power. 

Hello Ashish,

  As Alexey elude to above, I think you will need to use molecular dynamics or Monte Carlo method (maybe your best option) in order to search the very large conformational space of you peptide. This is especially important to do for a completely artificially build structure. If you are using the Linux OS, I would in general suggest using PyMol to build the peptide itself. However, most MD/MC programs will have preconstructed amino acid residues and a force-field that can be used with them. Depending on the program, it may or may not have the phosphate residue/parameters you are interested in. I know that xleap within the Amber program package will allow you to build a traditional peptide. Two other molecular mechanics-based programs are Gromacs and Charmm.

  If the program does not have the phosphate residue that you are interested in, or missing force-field parameters, then you will have to create them on your own. For amber there are several auxiliary programs to do this, including antechamber, R.E.D., and Wolf2Pack. If you are unfamiliar with MD/MC/force fields, this will require quite a bit of learn, but the modeling benefits will be many fold.

  You may also want to look into some very basic conformation generation algorithms (eg. Balloon Confab, OpenBabel). However, these are typically based on very rudimentary physics models, and I am not sure if they can handle such a large system as yours.

  Again, I think the key will be how you sample the conformational space of your system. Anyone who reviews your work will want to know how you did this, and how you justify the conformations that you end up reporting/comparing. Once you identify the possible conformations, you can always use a higher level of theory (e.g. quantum mechanics) for additional investigations.

Best regards,

Karl

The trouble with building molecules atom by atom using molecular dynamics or MC simulation is that they are reliant on too simple a model of nature.  The only way this is possible is by taking a more ab initio approach.  The most common is to use density functional theory (DFT) which is a way of treating all of the quantum effects that are present in reality but omitted by standard force fields.  Other methods like packages present in MM or MD packages tend to rely on more than just force fields and energy minimization and are therefore not doing strictly what it sounds to me you wish to do.

One additional thought is that to me it sounds like you are attempting to solve the protein folding problem.  Do you wish to just link atoms and then derive the protein structure.  If so, this is not yet a tenable problem.

  Here are a few more things to think about, and to read up on further if you will be pursuing this in depth.

  DFT, QM, Semiempirical, MM, and MD are all theories for modeling reality. Each have their own set of approximations, and subsequently they each introduce a level of inaccuracies into their calculations. Each also have their own strengths and own weaknesses (e.g. DFT is not variational consistent). QM, DFT, and semiempirical theories allow you to investigate bond breaking and formation, for example, that is not done with standard MM and MD. At the same time they are unable to sample a time regime that is accessible by MD simulations, nor do they account for the physiological environment (i.e. solvent, ions) explicitly. Plus, with a well parameterized force field, minimization of geometries can be competitive in terms of accuracy with some QM and DFT methods.

  The choice of which theory to use is dictated by the questions that you are asking. Thus, the questions are key to guiding your research design.

Molden or Avogadro are great. I like Molden because it is more handy. You may generate the geometry first and then optimize it using a suitable software.

I prefer to do it my self using pen and paper.

If you roughly know the bond-lengths, bond-angles and dihedral angles, generating a Z-matrix is not too difficult. From my experience with molecule-generating softwares, I find that it sometimes takes long time to generate the geometries when you have multiple rings in the structure. Instead, if approximate geometric parameters can be guessed, it is very convenient to generate the Z-matrix on paper and then type and create a geometry file, view on Molden or similar software, correct it if required and then optimize to get the optimum parameters. If you are not used to making the Z-matrix yourself, start with small examples, like benzene, naphthalene to get acquainted. Then, go for bigger molecules you are interested in.

Good luck!

There are many packages that will help you do this.

My favourite is Yasara. Just start with methane and build up from there. A molecular dynamics and semiemprical quantum methods are included so you can minimize your structures as you go. Easy ....

I suggest Avogadro. 

Please bear also in mind that if the molecule is very big their will be a lot of local minima. 

The best thing would be, after having built the molecule, to try a classical annealing procedure in order to select different structures, and then proceed to find quantistically the one that has the lower energy.

Thank you all for your valuable suggestions. Special thanks to Karl N Kirschner for explaining the concepts in such great detail and pointing me to literature that I need to read. I believe after I am done reading, I'll be able to design a good experiment.