Limits...
Replica-Exchange Accelerated Molecular Dynamics (REXAMD) Applied to Thermodynamic Integration.

Fajer M, Hamelberg D, McCammon JA - J Chem Theory Comput (2008)

Bottom Line: Accelerated molecular dynamics (AMD) is an efficient strategy for accelerating the sampling of molecular dynamics simulations, and observable quantities such as free energies derived on the biased AMD potential can be reweighted to yield results consistent with the original, unmodified potential.We propose a replica exchange of various degrees of acceleration (REXAMD) to retain good statistics while achieving enhanced sampling.The REXAMD method is validated and benchmarked on two simple gas-phase model systems, and two different strategies for computing reweighted averages over a simulation are compared.

View Article: PubMed Central - PubMed

Affiliation: Department of Chemistry and Biochemistry, University of California at San Diego, La Jolla, California 92093-0365, Center for Theoretical Biological Physics, University of California at San Diego, La Jolla, California 92039-0365, Department of Chemistry, Georgia State University, Atlanta, Georgia 30302-4098, Department of Pharmacology, University of California at San Diego, La Jolla, California 92093-0365, and Howard Hughes Medical Institute, University of California at San Diego, La Jolla, California 92093-0365.

ABSTRACT
Accelerated molecular dynamics (AMD) is an efficient strategy for accelerating the sampling of molecular dynamics simulations, and observable quantities such as free energies derived on the biased AMD potential can be reweighted to yield results consistent with the original, unmodified potential. In conventional AMD the reweighting procedure has an inherent statistical problem in systems with large acceleration, where the points with the largest biases will dominate the reweighted result and reduce the effective number of data points. We propose a replica exchange of various degrees of acceleration (REXAMD) to retain good statistics while achieving enhanced sampling. The REXAMD method is validated and benchmarked on two simple gas-phase model systems, and two different strategies for computing reweighted averages over a simulation are compared.

No MeSH data available.


Block average of the MSB thermodynamic integration results from the reweighted runs strategy shown on two different scales. The symbols show the average value of each simulation type, and the shaded region shows the standard error for each simulation type. The top plot shows the REXAMDtT-rw results on scale and shows how poor the statistics are after reweighting. The bottom plot shows the REXAMDtT results on scale and shows how quickly the REXAMD technique converges to within statistical accuracy.
© Copyright Policy - open-access - ccc-price
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC2651661&req=5

fig4: Block average of the MSB thermodynamic integration results from the reweighted runs strategy shown on two different scales. The symbols show the average value of each simulation type, and the shaded region shows the standard error for each simulation type. The top plot shows the REXAMDtT-rw results on scale and shows how poor the statistics are after reweighting. The bottom plot shows the REXAMDtT results on scale and shows how quickly the REXAMD technique converges to within statistical accuracy.

Mentions: The reweighting procedure was applied to the state with the highest degree of acceleration, s7, because this state is the most independent of the other states in terms of convergence. The most accelerated state is also expected to have the highest range of ΔV boost factors and therefore exhibit the largest reweighting problem.(8) This prediction can be seen in the poor accuracy and precision of the ΔG of reweighted runs for REXAMDtT-rw (Table 3, Figure 4). The effective numbers of data points for the s7 states are shown in Table S-I (Supporting Information) and demonstrate the source of the poor statistics. For example, the λ of 0.5 simulations had a standard deviation of boost values of 13 kcal/mol, and only 30 of the 200,000 data points from the ten duplicate runs contributed to ⟨dV/dλ⟩λ=0.5.


Replica-Exchange Accelerated Molecular Dynamics (REXAMD) Applied to Thermodynamic Integration.

Fajer M, Hamelberg D, McCammon JA - J Chem Theory Comput (2008)

Block average of the MSB thermodynamic integration results from the reweighted runs strategy shown on two different scales. The symbols show the average value of each simulation type, and the shaded region shows the standard error for each simulation type. The top plot shows the REXAMDtT-rw results on scale and shows how poor the statistics are after reweighting. The bottom plot shows the REXAMDtT results on scale and shows how quickly the REXAMD technique converges to within statistical accuracy.
© Copyright Policy - open-access - ccc-price
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC2651661&req=5

fig4: Block average of the MSB thermodynamic integration results from the reweighted runs strategy shown on two different scales. The symbols show the average value of each simulation type, and the shaded region shows the standard error for each simulation type. The top plot shows the REXAMDtT-rw results on scale and shows how poor the statistics are after reweighting. The bottom plot shows the REXAMDtT results on scale and shows how quickly the REXAMD technique converges to within statistical accuracy.
Mentions: The reweighting procedure was applied to the state with the highest degree of acceleration, s7, because this state is the most independent of the other states in terms of convergence. The most accelerated state is also expected to have the highest range of ΔV boost factors and therefore exhibit the largest reweighting problem.(8) This prediction can be seen in the poor accuracy and precision of the ΔG of reweighted runs for REXAMDtT-rw (Table 3, Figure 4). The effective numbers of data points for the s7 states are shown in Table S-I (Supporting Information) and demonstrate the source of the poor statistics. For example, the λ of 0.5 simulations had a standard deviation of boost values of 13 kcal/mol, and only 30 of the 200,000 data points from the ten duplicate runs contributed to ⟨dV/dλ⟩λ=0.5.

Bottom Line: Accelerated molecular dynamics (AMD) is an efficient strategy for accelerating the sampling of molecular dynamics simulations, and observable quantities such as free energies derived on the biased AMD potential can be reweighted to yield results consistent with the original, unmodified potential.We propose a replica exchange of various degrees of acceleration (REXAMD) to retain good statistics while achieving enhanced sampling.The REXAMD method is validated and benchmarked on two simple gas-phase model systems, and two different strategies for computing reweighted averages over a simulation are compared.

View Article: PubMed Central - PubMed

Affiliation: Department of Chemistry and Biochemistry, University of California at San Diego, La Jolla, California 92093-0365, Center for Theoretical Biological Physics, University of California at San Diego, La Jolla, California 92039-0365, Department of Chemistry, Georgia State University, Atlanta, Georgia 30302-4098, Department of Pharmacology, University of California at San Diego, La Jolla, California 92093-0365, and Howard Hughes Medical Institute, University of California at San Diego, La Jolla, California 92093-0365.

ABSTRACT
Accelerated molecular dynamics (AMD) is an efficient strategy for accelerating the sampling of molecular dynamics simulations, and observable quantities such as free energies derived on the biased AMD potential can be reweighted to yield results consistent with the original, unmodified potential. In conventional AMD the reweighting procedure has an inherent statistical problem in systems with large acceleration, where the points with the largest biases will dominate the reweighted result and reduce the effective number of data points. We propose a replica exchange of various degrees of acceleration (REXAMD) to retain good statistics while achieving enhanced sampling. The REXAMD method is validated and benchmarked on two simple gas-phase model systems, and two different strategies for computing reweighted averages over a simulation are compared.

No MeSH data available.