It mostly depends on the size of the protein and the domain structure. For a medium sized single domain protein, the ab initio modeling should give a good prediction. Prediction quality also varies depending on the type of the protein (membrane protein or cytosolic proteins etc.).

So far I know ab-initio modeling requires more time but it gives a good prediction.

Md. Abu Sufian

Could you please elaborate the answer.

There are three theories to calculate molecular mechanics (molecular geometries, reactivities,spectra, and other properties).

Ab initio methods – based on approximate solutions of the Schrodinger equation

without appeal to fitting to experiment. It considers all the motions in a molecule and which makes calculation with this theory time consuming.

Semiempirical methods – based on approximate solutions of the Schrodinger

equation with appeal to fitting to experiment (i.e. using parameterization). This method combines theory and experimental data to calculate structure.

Density functional theory (DFT) methods – based on approximate solutions of the

Schrodinger equation, bypassing the wavefunction that is a central feature of ab

initio and semiempirical methods.

That is why it is better to avoid ab initio method while calculating large structures. Semi empirical requires less time and DFT in between.

A PDF is also attached for further clarification.

Md. Abu Sufian

Thanks would be too shallow.

Uttam Pal 

Could you please tell me the reliability regarding the cytosolic proteins?

Here is a statistics for ROSETTA: http://www.sciencedirect.com/science/article/pii/S0022283600944590

You can look into that. It also analyzes the results based on different types of protein (according to SCOP classifications).

Thank You Uttam Pal