Machine-Learning Enabled New Insights into the Coil-to-Globule

6 days ago - (9−12) Various methods such as force-matching, the iterative Boltzmann approach, etc. have been utilized to develop CG models of these ...
0 downloads 0 Views 3MB Size
Download PDF
Letter pubs.acs.org/JPCL

Cite This: J. Phys. Chem. Lett. 2018, 9, 6480−6488

Machine-Learning Enabled New Insights into the Coil-to-Globule Transition of Thermosensitive Polymers Using a Coarse-Grained Model Karteek K. Bejagam,† Yaxin An,† Samrendra Singh,‡ and Sanket A. Deshmukh*,† †

Department of Chemical Engineering, Virginia Tech, Blacksburg, Virginia 24061, United States CNH Industrial, Burr Ridge, Illinois 60527, United States

‡

J. Phys. Chem. Lett. 2018.9:6480-6488. Downloaded from pubs.acs.org by UNIV OF NORTH DAKOTA on 11/18/18. For personal use only.

S Supporting Information *

ABSTRACT: We present a computational framework that integrates coarse-grained (CG) molecular dynamics (MD) simulations and a data-driven machine-learning (ML) method to gain insights into the conformations of polymers in solutions. We employ this framework to study conformational transition of a model thermosensitive polymer, poly(N-isopropylacrylamide) (PNIPAM). Here, we have developed the first of its kind, a temperatureindependent CG model of PNIPAM that can accurately predict its experimental lower critical solution temperature (LCST) while retaining the tacticity in the presence of an explicit water model. The CG model was extensively validated by performing CG MD simulations with different initial conformations, varying the radius of gyration of chain, the chain length, and the angle between the adjacent monomers of the initial configuration of PNIPAM (total simulation time = 90 μs). Moreover, for the first time, we utilize the nonmetric multidimensional scaling (NMDS) method, a data-driven ML approach, to gain further insights into the mechanisms and pathways of this coil-to-globule transition by analyzing CG MD simulation trajectories. NMDS analysis provides entirely new insights and shows multiple metastable states of PNIPAM during its coil-to-globule transition above the LCST.

T

thermosensitive polymers such as poly(N-isopropylacrylamide), poly(ethylene glycol), etc.20−22 In recent years, a range of different machine-learning (ML) methods such as deep learning, manifold learning, etc. have been utilized to gain insights into crystal structures, folding pathways for biomolecules, and aggregation of small organic molecules.23−26 A nonlinear manifold learning is related to the algorithmic techniques of dimensionality reduction, and it consists of different methods including diffusion maps, the nonmetric multidimensional scaling (NMDS) method, isomaps, Hessian Eigen maps, etc.27 The NMDS is a data-driven ML scheme, and it has been employed to reduce the high dimensions in the MD trajectories of protein folding while preserving the crucial features.25 Use of such a manifold learning method to analyze the CG MD simulation trajectories of stimuli-sensitive polymers may allow one to capture the rare events, if any, occurring during their conformational transitions. Thus, it can provide insights into the mechanism and pathways of conformational transition. Poly(N-isopropylacrylamide) (PNIPAM), a model thermosensitive polymer, exhibits a coil-to-globule conformational transition above its lower critical solution temperature (LCST)

he coarse-grained (CG) molecular dynamics (MD) simulations have emerged as a powerful technique to model polymers, biomolecules, and various architectures of these materials.1−8 CG models that represent a group of atoms as beads allow the use of a higher time step (5 to 50 fs) as compared to the all-atom (AA) MD simulations (1 to 2 fs). This enables us to reach the time- and length-scales of several tens of microseconds (μs) and tens of nanometers (nm), respectively.9−12 Various methods such as force-matching, the iterative Boltzmann approach, etc. have been utilized to develop CG models of these macromolecules and solvents.11,13−17 Recently, we have developed a novel approach that integrates an artificial neural network (ANN) based machine-learning model and particle swarm optimization (PSO) algorithm with MD simulations to accelerate the development of CG models (ANN-assisted PSO method).18 Despite all these advancements, developing CG models of any stimuli-sensitive polymers that can accurately capture their conformational transitions has been a grand challenge. This is mainly because capturing the delicate changes in the inter- and intra-molecular interactions between polymer and solvent, in response to change in the surrounding environment, is very difficult.19 A CG model that can capture these interactions, if developed, can be used to study the architectures and selfassembly processes in stimuli-sensitive polymers. Indeed, there have been a few recent efforts made to develop CG models of © 2018 American Chemical Society

Received: September 25, 2018 Accepted: October 29, 2018 Published: October 29, 2018 6480

DOI: 10.1021/acs.jpclett.8b02956 J. Phys. Chem. Lett. 2018, 9, 6480−6488

Letter

The Journal of Physical Chemistry Letters

Figure 1. (a) Mapping scheme of CG NIPAM model. (b) Schematic of CG models of PNIPAM and 1-site water.35 (c) Distribution of Rg, averaged over five independent simulations, for 30-mer at 290 K (below LCST, black lines) and 320 K (above LCST, red lines). Each CG MD simulation was performed for 1 μs in the NPT ensemble, and the histograms were determined for the last 850 ns (850 ns × 5 = 4250 ns). Inset shows the representative snapshots for the coil-like (black chain) and globule-like (red chain) conformations. Distribution of Rg for PNIPAM chains with (d) 5-mer, (e) 18-mer, (f) 30-mer, and (g) 100-mer at 290 and 320 K.

of 305 K.28 As this coil-to-globule transition alters its physicochemical properties, it finds applications in the field of biomedicine, energy, coating, etc.29−31 Developing a CG model of PNIPAM that accurately predicts its LCST is very challenging, as the model should precisely capture both the polymer−polymer and polymer−water interactions. To the best of our knowledge, only three CG models of PNIPAM have been reported in literature.32−34 However, they have one or more of the following drawbacks: (i) Although they show a coil-like and a globule-like state below and above the LCST of PNIPAM, respectively, the exact LCST of these models is not reported.32 (ii) They are temperature-dependent models; i.e., a set of force-field (FF) parameters used to define the interactions between polymer and water, below and above the LCST, are different.33 (iii) These models could not retain the chain tacticity. Note, the tacticity of PNIPAM is known to play an important role in determining the LCST of PNIPAM.34 Additionally, (iv) the CG MD simulations utilized the implicit water model.34 Hence, none of these models can capture the polymer−water and polymer−polymer interactions accurately.19

Here, we present a novel computational framework that integrates CG MD simulations and data-driven machinelearning methods to gain insights into conformations of polymers in solution. We employ this framework to study the coil-to-globule transition in PNIPAM. We began by developing the first of its kind temperature-independent, chemically and structurally accurate, transferable CG model of PNIPAM by coupling MD simulations with the PSO algorithm.18,35−37 We show that the PNIPAM CG model retains its tacticity and is able to reproduce its experimentally reported LCST behavior. Then, we employ NMDS method to analyze the CG MD simulation trajectories and to explore the mechanism and pathways of the coil-to-globule transition in PNIPAM.25 The NMDS analysis shows multiple metastable states of PNIPAM, which has not been reported for any thermosensitive polymer. The first step of the CG model development of PNIPAM involved the FF optimization of its monomer, N-isopropylacrylamide (NIPAM). The FF parameters were optimized to reproduce the properties obtained from the AA MD simulations. The development of the NIPAM CG model was based on the features and properties obtained from the AA 6481

DOI: 10.1021/acs.jpclett.8b02956 J. Phys. Chem. Lett. 2018, 9, 6480−6488

Letter

The Journal of Physical Chemistry Letters

respectively. The nonbonded interactions between PNIPAM chain and water beads (W1) were represented by 12-6 Lennard−Jones (LJ) potentials. The values of ε, the strength of interactions, between C2M-W1, AMP-W1, and ISP-W1 beads were optimized to reproduce the coil-like and globule-like conformation of PNIPAM at 290 K (below LCST) and 320 K (above LCST) (see Section 2 of the Supporting Information). The value of σ, the finite distance at which the intermolecular potential between two beads becomes zero, was obtained by the Lorentz−Berthelot (LB) mixing rule. Previous AA MD simulations of PNIPAM have reported radius of gyration (Rg) values of ∼16 Å and ∼10 Å for PNIPAM 30-mer at ∼290 K and ∼320 K, respectively.19 The Rg values of ∼16 Å and ∼10 Å represent a coil-like and a globule-like state of the PNIPAM 30-mer chain. In the present study, to optimize the FF parameters of PNIPAM, by using the PSO method, we set its Rg’s to ∼16 Å (at 290 K) and ∼10 Å (320 K) as target values.19 The FF parameters were optimized by performing simulations at 290 and 320 K temperatures simultaneously. The optimized FF parameters for the PNIPAM CG model are listed in Table S3 of the Supporting Information. The robustness of these FF parameters was evaluated by performing five independent simulations for 1 μs at 290 and 320 K. The distribution of Rg obtained by averaging the last 850 ns simulation trajectory from each of these simulations (850 ns × 5 = 4250 ns) is shown in Figure 1c. This figure clearly shows a coil-like (Rg ∼16 Å) and globule-like conformations (Rg ∼11 Å) of 30-mer of PNIPAM below and above its LCST, respectively. Furthermore, to evaluate the ability of the new PNIPAM CG model in capturing the dependence of LCST on the PNIPAM chain length, we have generated PNIPAM chains with 5-, 18-, and 100-mer. Note, both AA MD and experiments have shown that the PNIPAM chains below 10-mer do not exhibit a coil-to-globule transition due to their short chain length.19,44−46 Three different initial configurations were generated for each chain length to perform simulations in the presence of the 1-site CG water model. The simulations were carried out for 1 μs at 290 and 320 K, which is below and above the LCST of PNIPAM, respectively. Average Rg distributions obtained from both the simulations of different lengths are given in Figure 1d to 1g. Consistent with the previous reported literature, the 5-mer PNIPAM chain does not exhibit a coil-to-globule transition above its LCST.19,44,46 The peak position at both temperatures fluctuates at ∼5 Å. As the chain length is increased from 5mer to 18-mer, the PNIPAM chains exhibit a coil-like and globule-like state, below and above its LCST, respectively (Figure 1e). In the case of 30-mer and 100-mer at 290 K, there is clear evidence of a coil-like state of PNIPAM with Rg spreading across 13 Å to 19 Å and 30 Å to 60 Å, respectively. At 320 K, the peaks in Rg distributions for 30-mer and 100-mer were observed at ∼11 Å and ∼18 Å, respectively. Thus, the new CG PNIPAM model is able to capture the coil-to-globule transition for PNIPAM chains with different lengths. To evaluate the structure of water near PNIPAM chains, we calculated the RDF between 30-mer or 5-mer of PNIPAM and water beads, at two temperatures (see Supporting Information Figure S5). The PNIPAM chain with 30-mer dehydrates above the LCST which can be attributed to its globule-like state, consistent with AA results.47,48 Ahmed et al. have predicted the dehydration time for water within 1 μs using kinetic studies on the PNIPAM collapse, and our CG MD simulations of

NIPAM simulations that are modeled using CHARMM FF parameters.38 The AA MD simulations of NIPAM molecules was carried out at 300 K (see Supporting Information Section 1.1). The CG NIPAM model was mapped on its functional groups, which resulted in a model with three beads, namely C2M, AMP, and ISP, that encompass the CH2−CH2, H−N− CO, and isopropyl groups of a NIPAM monomer (Figure 1a). Beads connected with bonds interact with harmonic bond potentials (see Supporting Information Section 1.2). Beads separated by two bonds interact via harmonic angle potential as well as through nonbonded interactions. All three CG beads in NIPAM are chargeless and interact with each other via 12-6 Lennard−Jones (LJ) nonbonded potential. The AA simulation trajectory was converted into a CG trajectory based on the mapping scheme illustrated in Figure 1a. The equilibrium bonds and angle between different CG beads were determined based on the mean value of the distributions obtained from the mapped AA trajectories. The values of σ for AMP and ISP beads were chosen based on the radial distribution functions (RDFs) obtained from mapped AA trajectory. Nonbonded interaction parameters of C2M were adopted from our recently developed CG hydrocarbon model.36 Thus, to develop a CG model of NIPAM, we had to optimize the following five FF parameters: kb (for C2M-AMP; AMP-ISP), kθ (C2M-AMPISP), εAMP, and εISP. These FF parameters were optimized to reproduce the bonded distributions, RDFs obtained from the mapped trajectory, and its properties such as density, surface tension, and heat of vaporization obtained from NIPAM AA MD simulations. The optimization of FF parameters was performed by coupling the PSO method with MD simulations. The time step of 15 fs was used to perform the CG MD simulations at 300 K by using the NAMD simulation package.39 The optimized FF parameters for NIPAM CG model are listed in Table S1 of the Supporting Information. The properties obtained by using this CG model were in good agreement with those obtained from AA MD simulations (see Table S2 of the Supporting Information). The second step involved the development of intramolecular interactions for a PNIPAM chain and intermolecular interactions between PNIPAM and water to reproduce its LCST behavior. The chemically and structurally accurate NIPAM model was utilized to generate a PNIPAM chain with 30 monomer units (30-mer), which was obtained by placing adjacent monomers at a random angle. A 30-mer PNIPAM chain has been successfully studied to investigate the mechanism of coil-to-globule transition in many AA MD simulations by employing various FF parameters.19,40,41 The 30-mer CG PNIPAM chain was solvated with 8434 beads of the 1-site CG water model (see Figure 1b).35 Note, the 1-site CG model can reproduce several properties of water at various temperatures from 280 to 370 K.35 The bonded and nonbonded interaction parameters for the backbone of PNIPAM, represented by −C2M−C2M− beads, were adopted from the recently developed hydrocarbon CG model.36 The dihedral potentials were incorporated into the FF form with the goal of retaining the tacticity of PNIPAM chains (see Supporting Information Section 2 for more details).42,43 The equilibrium angle (C2M−C2M−AMP) and equilibrium dihedral angles were chosen based on the distributions obtained from the mapped AA trajectory of PNIPAM simulations. During the FF parametrization, angle force constants involved with backbone atoms and dihedral force constants are set to 15 kcal/mol/radian2 and 1.0 kcal/mol, 6482

DOI: 10.1021/acs.jpclett.8b02956 J. Phys. Chem. Lett. 2018, 9, 6480−6488

Letter

The Journal of Physical Chemistry Letters

Figure 2. Distribution of Rg profiles obtained by averaging the 10 different initial configurations at two temperatures: 290 K (below LCST, black lines) and 320 K (above LCST, red lines). All these ten configurations are selected based on their similar initial Rg values ranging from (a) 10−12 Å (complete globule-like state) (S-I), (b) 12−14 Å (partial globule-like state) (S-II), (c) 14−16 Å (partial coil-like state) (S-III), and (d) 16−18 Å (extended coil-like state) (S-IV). Representative snapshots for each set are shown as inset to each figure.

Figure 3. Evaluation of the LCST of three selected conformations from the initial Rg ranging between 14 and 18 Å: (a) P1LCST = ∼300 K, (b) P2LCST = ∼307 K, (c) P3LCST = ∼312 K. (d) Autocorrelation function of the dihedral angle (AMP-C2M-C2M-AMP), averaged over all the dihedrals except the edge ones. Simulation trajectories that corresponds to 290 and 320 K are shown with solid and dashed lines, respectively. (e) Distance distribution between the alternate ISP beads for the three conformations. Color scheme used to represent the data for P1, P2, and P3 are black, red, and blue, respectively.

6483

DOI: 10.1021/acs.jpclett.8b02956 J. Phys. Chem. Lett. 2018, 9, 6480−6488

Letter

The Journal of Physical Chemistry Letters

Figure 4. Distribution of Rg for PNIPAM chains with initial angle between adjacent NIPAM monomers of (a) 0°, (b) 30°, (c) 60°, (d) 90°, (e) 120°, (f) 150°, and (g) 180° at 290 and 320 K. All the profiles were averaged over three different initial configurations for a given angle, and the corresponding values for each simulation are given in Table S4 of the Supporting Information. Inset shows one of the initial configurations. (h) Distribution of the distance between the alternate ISP beads of various PNIPAM conformations with different initial angles between the NIPAM monomers.

PNIPAM, respectively. These results are consistent with AA simulations reported by Kang et al.41 This further suggests that the CG model of PNIPAM, developed in this study, can accurately capture the behavior reported for AA models of PNIPAM. Furthermore, to investigate the ability of the CG model in predicting the LCST of PNIPAM, we randomly selected three configurations (P1, P2, and P3) that showed an LCST transition and whose Rg ranged from 14 to 18 Å (sets S-III and S-IV). If the interactions between PNIPAM and water are captured accurately, then the LCST predicted by the new CG model should fall within the range observed in experiments.49 Note, it has been reported that the tacticity of PNIPAM plays an important role in determining its LCST, and based on its tacticity, the LCST can vary from 290 to 305 K.49 For example, the PNIPAM chains with a meso-diad content of ∼45% and ∼66% were known to show LCST at ∼305 K and ∼290 K, respectively.19,49 Here, the CG MD simulations were performed for 1 μs for temperatures ranging from 285 to 320 K with an interval of 5 K. Figure 3a−c shows the mean Rg value over the last 850 ns and its standard deviation at each temperature for the PNIPAM with three different initial conformations. The LCST predicted by the three different models is ∼300 K (P1), ∼307 K (P2), and ∼312 K (P3), which is comparable to the experimentally reported LCST of PNIPAM (290 to 305 K). The differences in these LCSTs suggest that the three chains may have different tacticities. Therefore, the three simulated chains can be attributed to a meso-diad content of ∼57% (P1LCST = ∼300 K), ∼45% (P2LCST = ∼307 K), and less than 45% (P3LCST = ∼312 K) based on the experimental study.49 Recently, Boţan et al. used an autocorrelation function (ACF) of a dihedral angle (ACF = ⟨cos(Θ(t) − Θ(t + dτ))⟩, with Θ being the dihedral angle at time t to demonstrate that, for an AA model of PNIPAM chain, the tacticity is retained.34 Moreover, they also show that retaining such tacticity in a CG model is a very challenging task. To evaluate the ability of the

PNIPAM chains (see Figure S6 of the Supporting Information) suggest comparable time scales with this study.45 Experimentally, the coil-to-globule transition exhibited by PNIPAM is reversible; i.e., when the temperature is decreased from above its LCST to below the LCST, PNIPAM undergoes a globule-to-coil transition. However, Kang et al. have shown that the tightly collapsed initial configurations (a globule-like state) remain in the globule-like state below its LCST during 1 μs AA MD simulations, which suggests that these time scales could be short to observe its globule-to-coil transition.41 To further explore the robustness of our new CG model of PNIPAM in predicting such behavior, 40 different configurations of PNIPAM with initial Rg values ranging from 10 Å to 18 Å were generated. These 40 structures were binned based on their Rg’s such that every 10 configurations fall in the following sets: 10−12 Å (complete globule-like state) (S-I), 12−14 Å (partial globule-like state) (S-II), 14−16 Å (partial coil-like state) (S-III), and 16−18 Å (extended coil-like state) (S-IV). MD simulations of all the PNIPAM structures were carried out for 1 μs in the NPT ensemble at 290 K (below LCST) and 320 K (above LCST). The Rg value for an individual conformation was determined over the last 850 ns of the total trajectory. The data obtained for individual conformations are shown in Figure S6 of the Supporting Information. Figure 2a−d shows the average distribution of Rg obtained from each set. Overall, we find that with the increase in the initial Rg values, an increasing number of PNIPAM chains showed a coil-like and a globule-like state, below and above the LCST of PNIPAM, respectively. In Figure 2a−d, at 290 K, the distributions are bimodal in nature since they represent the average data for 10 configurations irrespective of whether they were in a coil-like or a globule-like state. For S-I and S-II, the average Rg values suggest the presence of PNIPAM chains in a globule-like state both below and above the LCST. In contrast, the distributions suggests that the conformations of S-III and S-IV demonstrate a clear coil-like and a globule-like state below and above the LCST of 6484

DOI: 10.1021/acs.jpclett.8b02956 J. Phys. Chem. Lett. 2018, 9, 6480−6488

Letter

The Journal of Physical Chemistry Letters

Figure 5. NMDS analysis showing the clustering, based on the similarity between structures in the dihedral space. (a) Evolution of a PNIPAM chain with an angle of 90° between adjacent NIPAM monomers, at two temperatures (290 and 320 K), illustrating the LCST behavior. (b) Different pathways opted by PNIPAM chains in reaching the globule-like state at 320 K are shown for a PNIPAM chain with different backbone configurations and an angle of 90° between adjacent NIPAM monomers. Colors of the points indicate the Rg value in accordance to the provided scale bar. Black and red color chains represent the snapshots that correspond to the simulations of 290 and 320 K, respectively. Initial PNIPAM conformation is shown in green color. Time evolution is indicated by dashed line with arrow to guide the eye.

290 and 320 K for 1 μs. The 1-site CG water model was used to solvate a PNIPAM chain. The Rg for each chain was evaluated for the final 850 ns of the simulation trajectory. As shown in Figure 4a−g, a coil-to-globule transition is observed for PNIPAM chains with initial angles of 90°, 120°, and 150°. The mean values of Rg are ∼18 Å and ∼11 Å at 290 and 320 K, respectively. For structures with an angle of 0°, 30°, 60°, and 180°, the PNIPAM chains were in the globule-like conformation (mean Rg = ∼12 Å) at both temperatures. To gain more insights into the observed behavior, distances between the ISP beads of alternate monomers were determined for the first 150 ns from the simulation trajectories. Figure 4h illustrates the distance distribution (averaged over three different simulations) between the ISP beads of alternate monomers for the LCST (90°, 120°, and 150°) and non-LCST (0°, 30°, 60°, and 180°) PNIPAM conformations. Overall, for the conformations that showed LCST behavior, the peak height at ∼5 Å and ∼11.5 Å increases and decreases, respectively, as compared to the non-LCST conformations. This may enhance the polymer−polymer interactions for nonLCST conformations, and thereby PNIPAM chains form a globule-like structure below and above the LCST. The structure of water around the AMP beads of PNIPAM chains for the LCST ones (90°, 120°, and 150°) were determined at two temperatures (see Figure S8 of the Supporting Information). RDF profiles suggest that the dehydration time scales for amide groups (3) during its coil-toglobule transition (see Supporting Information Section 5 and Figure S9). This may allow them to retain a coil-like conformation below LCST and undergo a coil-to-globule transition above the LCST. In summary, we present a novel computational framework that integrates coarse-grained (CG) molecular dynamics (MD) simulations with a data-driven machine-learning (ML) method, which we use to study a classic thermosensitive polymer, poly(N-isopropylacrylamide) (PNIPAM). We have developed an accurate temperature-independent CG model of PNIPAM by integrating the particle swarm optimization (PSO) method with MD simulations. This is a first of its kind CG model of PNIPAM that can retain its tacticity and accurately reproduce the dependence of LCST on its tacticity. This model was thoroughly validated by predicting the LCST of PNIPAM chains with different initial conformations of the chain (different Rg as well as different angle between the adjacent monomers), with varied chain length in the presence of the 1-site CG water model. The analyses of 90 μs simulation trajectories clearly suggest that the new CG model could accurately reproduce the experimentally reported value of the LCST of ∼305 K. Moreover, to gain insights into the pathways of the coil-to-globule transition of PNIPAM, we have utilized the nonmetric multidimensional scaling (NMDS) method, a data-driven ML approach, to analyze these simulation trajectories. To the best of our knowledge, this is the first study that uses an ML method to analyze the CG MD simulation trajectories of a thermosensitive polymer. This

analysis reveals that the PNIPAM chains exhibiting a coil-like and a globule-like state, below and above the LCST, respectively, go through multiple metastable states. Also, a coil-to-globule transition can occur through several different pathways. This knowledge can be utilized to systematically develop a thermosensitive polymer with a desired LCST temperature and transition pathways. Moreover, the use of ML methods to unravel pathways of coil-to-globule transition demonstrates their use in studying complex systems of stimulisensitive polymers. This new computational framework is general and can be utilized to study complex processes such as self-assembly and conformational transitions in the architectures of stimuli-sensitive polymers.

■

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jpclett.8b02956.

■

Details on the NIPAM and PNIPAM model development; force field parameters for the developed PNIPAM CG model; discussion of the structure of water around PNIPAM chains at two temperatures; more details on NMDS analysis (PDF)

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Phone: +1 540-231-8785. ORCID

Karteek K. Bejagam: 0000-0002-8660-9946 Sanket A. Deshmukh: 0000-0001-7573-0057 Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS Authors would like to acknowledge Advanced Research Computing (ARC), Virginia Tech for providing the computational resources. This research used resources of the National Energy Research Scientific Computing Center (NERSC), a U.S. Department of Energy Office of Science User Facility operated under Contract No. DE-AC02-05CH11231. S.A.D. acknowledges the Start-up Funds from Virginia Tech.

■

REFERENCES

(1) Marrink, S. J.; Risselada, H. J.; Yefimov, S.; Tieleman, D. P.; de Vries, A. H. The MARTINI Force Field: Coarse Grained Model for Biomolecular Simulations. J. Phys. Chem. B 2007, 111, 7812−7824. (2) Kenzaki, H.; Koga, N.; Hori, N.; Kanada, R.; Li, W.; Okazaki, K.I.; Yao, X.-Q.; Takada, S. CafeMol: A Coarse-Grained Biomolecular Simulator for Simulating Proteins at Work. J. Chem. Theory Comput. 2011, 7, 1979−1989. (3) Rossi, G.; Monticelli, L.; Puisto, S. R.; Vattulainen, I.; Ala-Nissila, T. Coarse-Graining Polymers with the MARTINI Force-Field: Polystyrene as a Benchmark Case. Soft Matter 2011, 7, 698−708. (4) Carbone, P.; Varzaneh, H. A. K.; Chen, X.; Müller-Plathe, F. Transferability of Coarse-Grained Force Fields: The Polymer Case. J. Chem. Phys. 2008, 128, 064904. (5) Deshmukh, S. A.; Solomon, L. A.; Kamath, G.; Fry, H. C. Water Ordering Controls the Dynamic Equilibrium of Micelle−fibre Formation in Self-Assembly of Peptide Amphiphiles. Nat. Commun. 2016, 7, 12367. 6486

DOI: 10.1021/acs.jpclett.8b02956 J. Phys. Chem. Lett. 2018, 9, 6480−6488

Letter

The Journal of Physical Chemistry Letters

(27) Izenman, A. J. Introduction to Manifold Learning. WIREs Comp Stat 2012, 4, 439−446. (28) Jain, K.; Vedarajan, R.; Watanabe, M.; Ishikiriyama, M.; Matsumi, N. Tunable LCST Behavior of poly(N-Isopropylacrylamide/ionic Liquid) Copolymers. Polym. Chem. 2015, 6, 6819−6825. (29) Guan, Y.; Zhang, Y. PNIPAM Microgels for Biomedical Applications: From Dispersed Particles to 3D Assemblies. Soft Matter 2011, 7, 6375−6384. (30) Gandhi, A.; Paul, A.; Sen, S. O.; Sen, K. K. Studies on Thermoresponsive Polymers: Phase Behaviour, Drug Delivery and Biomedical Applications. Asian J. Pharm. Sci. 2015, 10, 99−107. (31) Zhernenkov, M.; Ashkar, R.; Feng, H.; Akintewe, O. O.; Gallant, N. D.; Toomey, R.; Ankner, J. F.; Pynn, R. Thermoresponsive PNIPAM Coatings on Nanostructured Gratings for Cell Alignment and Release. ACS Appl. Mater. Interfaces 2015, 7, 11857−11862. (32) Wang, Z.; He, X. Developing Coarse-Grained Force Fields for PNIPAM Single Chain from the Atomistic Model. In 2009 3rd International Conference on Bioinformatics and Biomedical Engineering; 2009; pp 1−4. (33) Abbott, L. J.; Stevens, M. J. A Temperature-Dependent CoarseGrained Model for the Thermoresponsive Polymer poly(NIsopropylacrylamide). J. Chem. Phys. 2015, 143, 244901. (34) Boţan, V.; Ustach, V. D.; Leonhard, K.; Faller, R. Development and Application of a Coarse-Grained Model for PNIPAM by Iterative Boltzmann Inversion and Its Combination with Lattice Boltzmann Hydrodynamics. J. Phys. Chem. B 2017, 121, 10394−10406. (35) Bejagam, K. K.; Singh, S.; An, Y.; Berry, C.; Deshmukh, S. A. PSO-Assisted Development of New Transferable Coarse-Grained Water Models. J. Phys. Chem. B 2018, 122, 1958−1971. (36) An, Y.; Bejagam, K. K.; Deshmukh, S. A. Development of New Transferable Coarse-Grained Models of Hydrocarbons. J. Phys. Chem. B 2018, 122, 7143−7153. (37) Bejagam, K. K.; Singh, S.; Deshmukh, S. A. Development of Non-Bonded Interaction Parameters between Graphene and Water Using Particle Swarm Optimization. J. Comput. Chem. 2018, 39, 721− 734. (38) Vanommeslaeghe, K.; Hatcher, E.; Acharya, C.; Kundu, S.; Zhong, S.; Shim, J.; Darian, E.; Guvench, O.; Lopes, P.; Vorobyov, I.; et al. CHARMM General Force Field: A Force Field for Drug-like Molecules Compatible with the CHARMM All-Atom Additive Biological Force Fields. J. Comput. Chem. 2010, 31, 671−690. (39) Phillips, J. C.; Braun, R.; Wang, W.; Gumbart, J.; Tajkhorshid, E.; Villa, E.; Chipot, C.; Skeel, R. D.; Kalé, L.; Schulten, K. Scalable Molecular Dynamics with NAMD. J. Comput. Chem. 2005, 26, 1781− 1802. (40) Chiessi, E.; Paradossi, G. Influence of Tacticity on Hydrophobicity of Poly(N-Isopropylacrylamide): A Single Chain Molecular Dynamics Simulation Study. J. Phys. Chem. B 2016, 120, 3765−3776. (41) Kang, Y.; Joo, H.; Kim, J. S. Collapse−Swelling Transitions of a Thermoresponsive, Single Poly(N-Isopropylacrylamide) Chain in Water. J. Phys. Chem. B 2016, 120, 13184−13192. (42) Harmandaris, V. A.; Adhikari, N. P.; van der Vegt, N. F. A.; Kremer, K. Hierarchical Modeling of Polystyrene: From Atomistic to Coarse-Grained Simulations. Macromolecules 2006, 39, 6708−6719. (43) Li, C.; Shen, J.; Peter, C.; van der Vegt, N. F. A. A Chemically Accurate Implicit-Solvent Coarse-Grained Model for Polystyrenesulfonate Solutions. Macromolecules 2012, 45, 2551−2561. (44) Abbott, L. J.; Tucker, A. K.; Stevens, M. J. Single Chain Structure of a poly(N-Isopropylacrylamide) Surfactant in Water. J. Phys. Chem. B 2015, 119, 3837−3845. (45) Ahmed, Z.; Gooding, E. A.; Pimenov, K. V.; Wang, L.; Asher, S. A. UV Resonance Raman Determination of Molecular Mechanism of poly(N-Isopropylacrylamide) Volume Phase Transition. J. Phys. Chem. B 2009, 113, 4248−4256. (46) Deshmukh, S. A.; Sankaranarayanan, S. K. R. S.; Mancini, D. C. Vibrational Spectra of Proximal Water in a Thermo-Sensitive Polymer Undergoing Conformational Transition across the Lower Critical Solution Temperature. J. Phys. Chem. B 2012, 116, 5501−5515.

(6) Srinivas, G.; Discher, D. E.; Klein, M. L. Self-Assembly and Properties of Diblock Copolymers by Coarse-Grain Molecular Dynamics. Nat. Mater. 2004, 3, 638−644. (7) Yesylevskyy, S. O.; Schäfer, L. V.; Sengupta, D.; Marrink, S. J. Polarizable Water Model for the Coarse-Grained MARTINI Force Field. PLoS Comput. Biol. 2010, 6, e1000810. (8) Sharma, S.; Kumar, S. K.; Buldyrev, S. V.; Debenedetti, P. G.; Rossky, P. J.; Stanley, H. E. A Coarse-Grained Protein Model in a Water-like Solvent. Sci. Rep. 2013, 3, 1841. (9) Horn, J. N.; Sengillo, J. D.; Lin, D.; Romo, T. D.; Grossfield, A. Characterization of a Potent Antimicrobial Lipopeptide via CoarseGrained Molecular Dynamics. Biochim. Biophys. Acta, Biomembr. 2012, 1818, 212−218. (10) Hong, C.; Tieleman, D. P.; Wang, Y. Microsecond Molecular Dynamics Simulations of Lipid Mixing. Langmuir 2014, 30, 11993− 12001. (11) Lee, H.; de Vries, A. H.; Marrink, S.-J.; Pastor, R. W. A CoarseGrained Model for Polyethylene Oxide and Polyethylene Glycol: Conformation and Hydrodynamics. J. Phys. Chem. B 2009, 113, 13186−13194. (12) Winter, N. D.; Schatz, G. C. Coarse-Grained Molecular Dynamics Study of Permeability Enhancement in DPPC Bilayers by Incorporation of Lysolipid. J. Phys. Chem. B 2010, 114, 5053−5060. (13) Banerjee, P.; Roy, S.; Nair, N. Coarse-Grained Molecular Dynamics Force-Field for Polyacrylamide in Infinite Dilution Derived from Iterative Boltzmann Inversion and MARTINI Force-Field. J. Phys. Chem. B 2018, 122, 1516−1524. (14) Lu, L.; Dama, J. F.; Voth, G. A. Fitting Coarse-Grained Distribution Functions through an Iterative Force-Matching Method. J. Chem. Phys. 2013, 139, 121906. (15) Wang, Q.; Keffer, D. J.; Nicholson, D. M. A Coarse-Grained Model for Polyethylene Glycol Polymer. J. Chem. Phys. 2011, 135, 214903. (16) Prasitnok, K. A Coarse-Grained Model for Polylactide: Glass Transition Temperature and Conformational Properties. J. Polym. Res. 2016, 23, 139. (17) Peter, E. K.; Shea, J.-E.; Pivkin, I. V. Coarse kMC-Based Replica Exchange Algorithms for the Accelerated Simulation of Protein Folding in Explicit Solvent. Phys. Chem. Chem. Phys. 2016, 18, 13052−13065. (18) Bejagam, K. K.; Singh, S. K.; An, Y.; Deshmukh, S. A. Machine Learnt Coarse-Grained Models. J. Phys. Chem. Lett. 2018, 9, 4667− 4672. (19) Deshmukh, S. A.; Sankaranarayanan, S. K. R. S.; Suthar, K.; Mancini, D. C. Role of Solvation Dynamics and Local Ordering of Water in Inducing Conformational Transitions in poly(N-Isopropylacrylamide) Oligomers through the LCST. J. Phys. Chem. B 2012, 116, 2651−2663. (20) De Silva, C. C.; Leophairatana, P.; Ohkuma, T.; Koberstein, J. T.; Kremer, K.; Mukherji, D. Sequence Transferable Coarse-Grained Model of Amphiphilic Copolymers. J. Chem. Phys. 2017, 147, 064904. (21) Chudoba, R.; Heyda, J.; Dzubiella, J. Temperature-Dependent Implicit-Solvent Model of Polyethylene Glycol in Aqueous Solution. J. Chem. Theory Comput. 2017, 13, 6317−6327. (22) Abbott, L. J.; Stevens, M. J. A Temperature-Dependent CoarseGrained Model for the Thermoresponsive Polymer poly(NIsopropylacrylamide). J. Chem. Phys. 2015, 143, 244901. (23) Park, W. B.; Chung, J.; Jung, J.; Sohn, K.; Singh, S. P.; Pyo, M.; Shin, N.; Sohn, K.-S. Classification of Crystal Structure Using a Convolutional Neural Network. IUCrJ 2017, 4, 486−494. (24) Ryan, K.; Lengyel, J.; Shatruk, M. Crystal Structure Prediction via Deep Learning. J. Am. Chem. Soc. 2018, 140, 10158−10168. (25) Rajan, A.; Freddolino, P. L.; Schulten, K. Going beyond Clustering in MD Trajectory Analysis: An Application to Villin Headpiece Folding. PLoS One 2010, 5, e9890. (26) Wang, J.; Gayatri, M.; Ferguson, A. L. Coarse-Grained Molecular Simulation and Nonlinear Manifold Learning of Archipelago Asphaltene Aggregation and Folding. J. Phys. Chem. B 2018, 122, 6627−6647. 6487

DOI: 10.1021/acs.jpclett.8b02956 J. Phys. Chem. Lett. 2018, 9, 6480−6488

Letter

The Journal of Physical Chemistry Letters (47) Deshmukh, S. A.; Li, Z.; Kamath, G.; Suthar, K. J.; Sankaranarayanan, S. K. R. S.; Mancini, D. C. Atomistic Insights into Solvation Dynamics and Conformational Transformation in Thermo-Sensitive and Non-Thermo-Sensitive Oligomers. Polymer 2013, 54, 210−222. (48) Futscher, M. H.; Philipp, M.; Müller-Buschbaum, P.; Schulte, A. The Role of Backbone Hydration of Poly(N-Isopropyl Acrylamide) Across the Volume Phase Transition Compared to Its Monomer. Sci. Rep. 2017, 7, 17012. (49) Ray, B.; Okamoto, Y.; Kamigaito, M.; Sawamoto, M.; Seno, K.I.; Kanaoka, S.; Aoshima, S. Effect of Tacticity of Poly (NIsopropylacrylamide) on the Phase Separation Temperature of Its Aqueous Solutions. Polym. J. 2005, 37, 234. (50) Taguchi, Y.-H.; Oono, Y. Nonmetric Multidimensional Scaling As a Data-Mining Tool: New Algorithm and New Targets: Applications to Chemical Reaction Dynamics in Complex Systems. In Geometric Structures of Phase Space in Multidimensional Chaos; Toda, M., Komatsuzaki, T., Konishi, T., Berry, R. S., Rice, S. A., Eds.; Advances in Chemical Physics; John Wiley & Sons, Inc.: Hoboken, NJ, USA, 2005; pp 315−351. (51) Pereda, M.; Estrada, E. Machine Learning Analysis of Complex Networks in Hyperspherical Space. arXiv [physics.soc-ph], 2018. (52) Ferguson, A. L.; Panagiotopoulos, A. Z.; Kevrekidis, I. G.; Debenedetti, P. G. Nonlinear Dimensionality Reduction in Molecular Simulation: The Diffusion Map Approach. Chem. Phys. Lett. 2011, 509, 1−11. (53) Ferguson, A. L.; Panagiotopoulos, A. Z.; Debenedetti, P. G.; Kevrekidis, I. G. Systematic Determination of Order Parameters for Chain Dynamics Using Diffusion Maps. Proc. Natl. Acad. Sci. U. S. A. 2010, 107, 13597−13602.

6488

DOI: 10.1021/acs.jpclett.8b02956 J. Phys. Chem. Lett. 2018, 9, 6480−6488