Computational simulation of biological systems: Studies on protein folding and protein structure prediction

Scientific understanding as well as the way of studying science has been greatly changed since the advent of computer modeling. Computer simulation has played a central role in bridging theoretical and experimental studies. In this work, computer simulations were applied to explore biological systems on both protein folding and protein structure prediction studies. In the first study, the folding mechanisms of two alanine based helical peptides (Fs-21 peptide and MABA bonded Fs-21 peptide) were investigated by all atom molecular dynamics simulations and compared with experimental results. Multi-phase folding processes were observed for both peptides. Temperature change affected the relative stability of different ensembles. Helix-turn-helix conformation was found to be the most populated state at 300K while the full helix became more stable at low temperature (273K). The turn structure was found to be stabilized mainly by hydrophobic interactions. In the second study, helix-coil transition theory was elaborately tested by both statistical and energetic methods based on simulations of alanine based peptides. A weighted Ising model was proposed, and the model-derived propagation constant agreed very well with the experimental results. Solvation effect and electrostatic interactions were found to be the two main contributors to helix-coil transition. The results challenged the classic helix-coil transition theory by proving that the single sequence assumption was not appropriate for helix-coil transition. Conformational sampling has been a long-standing issue in computational sciences. In the third study, we systematically tested the convergence of the Replica Exchange Molecular Dynamics method (REMD), which is a recently developed method for conformational sampling enhancement. The results suggested that REMD can significantly enhance the sampling efficiency and accurately reproduce the long-time MD results with high efficiency. However, fluctuations at low temperatures (<300 K) indicated that REMD simulations did not converge within our simulation time (14 ns). Much longer REMD simulation time might be needed for the system to reach thermodynamic equilibrium than expected. Finding the optimal side chain packing is a common issue in structure prediction, protein design and protein docking. In the fourth study, a new method was presented. The method overcame the rough energy landscape problem and enabled all-atom MD simulation to be applied directly to protein structure refinement. The method showed very successful results on buried side-chain assignments, nearly 100% accuracy on all 6 randomly picked proteins was reached; the results also clearly demonstrated that the proposed method can significantly enhance conformational sampling. These encouraging results suggested prospective applications on many other protein related systems.