Predicting the functional state of protein kinases using interpretable graph neural networks from sequence and structural data.

Protein kinases are central to cellular activities and are actively pursued as drug targets for several conditions including cancer and autoimmune diseases. Despite the availability of a large structural database for kinases, methodologies to elucidate the structure-function relationship of these proteins (without manual intervention) are lacking. Such techniques are essential in structural biology and to accelerate drug discovery efforts. Here, we implement an interpretable graph neural network (GNN) framework for classifying the functionally active and inactive states of a large set of protein kinases by only using their tertiary structure and amino acid sequence. We show that the GNN models can classify kinase structures with high accuracy (>97%). We implement the Gradient-weighted Class Activation Mapping for graphs (Graph Grad-CAM) to automatically identify structurally important residues and residue-residue contacts of the kinases without any a priori input. We show that the motifs identified through the Graph Grad-CAM methodology are functionally critical, consistent with the existing kinase literature. Notably, the highly conserved DFG and HRD motifs of the well-known hydrophobic spine are identified by the interpretable framework in addition to some of the lesser known motifs. Further, using Grad-CAM maps as the vector embedding of the protein structures, we identify the subtle differences in the crystal structures among different sub-classes of kinases in the Protein Data Bank (PDB). Frameworks such as the one implemented here, for high-throughput identification of protein structure-function relationships are essential in designing targeted small molecules therapies as well as in engineering new proteins for novel applications.

Relationship between the Relaxation of Ionic Liquid Structural Motifs and That of the Shear Viscosity.

In a set of recent articles, we have highlighted that friction is highly inhomogeneous in a typical ionic liquid (IL) with charge networks that are stiff and charge-depleted regions that are soft. This has consequences not only for the dynamics of ILs but also for the transport properties of solutes dissolved in them. In this article, we explore whether the family of alkylimidazolium ILs coupled with bis(trifluoromethylsulfonyl)imide (with similar Coulombic interactions but different alkyl tails), when dynamically "equalized" by having a similar shear viscosity, display q-dependent structural relaxation time scales that are the same across the family. Our results show that this is not the case, and in fact, the relaxation of in-network charge alternation appears to be significantly affected by the presence of separate polar and apolar domains. However, we also find that if one was to assign weight factors to the relaxation of the structural motifs, charge alternation always contributes about the same amount (between 62.1 and 66.3%) across systems to the running integral of the stress tensor correlation function from which the shear viscosity is derived. Adjacency correlations between positive and negative moieties also contribute an identical amount if a prepeak is not present (about 38%) and a slightly smaller amount (about 28%) when intermediate range order exists. The prepeak only contributes about 6% to viscoelastic relaxation, highlighting that the dynamics of the smaller scale motifs is the most important.

A Pictorial View of Viscosity in Ionic Liquids and the Link to Nanostructural Heterogeneity.

Prototypical ionic liquids (ILs) are characterized by three structural motifs associated with (1) vicinal interactions, (2) the formation of positive-negative charge-alternating chains or networks, and (3) the alternation of these networks with apolar domains. In recent articles, we highlighted that the friction and mobility in these systems are nowhere close to being spatially homogeneous. This results in what one could call mechanical heterogeneity, where charge networks are intrinsically stiff and charge-depleted regions are softer, flexible, and mobile. This Letter attempts to provide a clear and visual connection between friction-associated with the dynamics of the structural motifs (in particular, the charge network)-and recent theoretical work by Yamaguchi linking the time-dependent viscosity of ILs to the decay of the charge alternation peak in the dynamic structure function. We propose that charge blurring associated with the loss of memory of where positive and negative charges are within networks is the key mechanism associated with viscosity in ILs. An IL will have low viscosity if a characteristic charge-blurring decorrelation time is low. With this in mind, engineering new low-viscosity ILs is reduced to understanding how to minimize this quantity.

In an ionic liquid, high local friction is determined by the proximity to the charge network.

Structural heterogeneity in Ionic Liquids (ILs) is to a large extent defined by nanoscale apolar pockets that act as spacers between strings of positive and negative charges that alternate. In contrast to this, recent work from our group and that of others appear to indicate that dynamic, energetic, and mechanical heterogeneities are governed by the charged part of the liquid. In this article, we study the dynamics of methane, a small apolar solute, in the family of ILs 1-alkyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide ( Im 1 , n + /NTf2 -), with n = 2, 4, 8 at temperatures that make the viscosity for each liquid similar and around 8 cP. We do this in an attempt to equalize the effect of the solvent on the dynamics of the solute. In all cases, we find that solute proximity to charge-enhanced regions coincides with translationally caged regimes (high local friction) whereas the opposite is true in charge-depleted regions. In a way, these ILs behave like a liquid within a liquid where the charge network is the high friction component.

Communication: Stiff and soft nano-environments and the "Octopus Effect" are the crux of ionic liquid structural and dynamical heterogeneity.

In a recent set of articles [J. C. Araque et al., J. Phys. Chem. B 119(23), 7015-7029 (2015) and J. C. Araque et al., J. Chem. Phys. 144, 204504 (2016)], we proposed the idea that for small neutral and charged solutes dissolved in ionic liquids, deviation from simple hydrodynamic predictions in translational and rotational dynamics can be explained in terms of diffusion through nano-environments that are stiff (high electrostriction, charge density, and number density) and others that are soft (charge depleted). The current article takes a purely solvent-centric approach in trying to provide molecular detail and intuitive visual understanding of time-dependent local mobility focusing on the most common case of an ionic liquid with well defined polar and apolar nano-domains. We find that at intermediate time scales, apolar regions are fluid, whereas the charge network is much less mobile. Because apolar domains and cationic heads must diffuse as single species, at long time the difference in mobility also necessarily dissipates.

Rotational Dynamics in Ionic Liquids from NMR Relaxation Experiments and Simulations: Benzene and 1-Ethyl-3-Methylimidazolium.

Temperature-dependent (2)H longitudinal spin relaxation times (T1) of dilute benzene-d6 in 1-butyl-3-methylimidazolium tetrafluoroborate ([Im41][BF4]) and two deuterated variants of the 1-ethyl-3-methylimidazolium cation (Im21(+)-d1 and Im21(+)-d6) in 1-ethyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide ([Im21][Tf2N]), measured at multiple Larmor frequencies, were used to probe rotational dynamics in ionic liquids. Rotational correlation times significantly faster than predicted by slip hydrodynamic calculations were observed for both solutes. Molecular dynamics simulations of these systems enabled extraction of more information about the rotational dynamics from the NMR data than rotation times alone. The multifrequency (2)H T1(T) data could be fit to within uncertainties over a broad region about the T1 minimum using models of the relevant rotational time correlation functions and their viscosity/temperature dependence derived from simulation. Such simulation-guided fitting provided confidence in the semiquantitative accuracy of the simulation models and enabled interpretation of NMR measurements to higher viscosities than previously possible. Simulations of the benzene system were therefore used to explore the nature of solute rotation in ionic liquids and how it might differ from rotation in conventional solvents. Whereas "spinning" about the C6 axis of benzene senses similarly weak solvent friction in both types of solvents, "tumbling" (rotations about in-plane axes) differs significantly in conventional solvents and ionic liquids. In the sluggish environment provided by ionic liquids, orientational caging and the presence of rare but influential large-amplitude (180°) jumps about in-plane axes lead to rotations being markedly nondiffusive, especially below room temperature.

A link between structure, diffusion and rotations of hydrogen bonding tracers in ionic liquids.

When solutes are small compared to the size of the ions in an ionic liquid, energetic heterogeneities associated with charge enhanced (stiff) and charge depleted (soft) nanoenvironments are sampled. In a recent article [J. C. Araque et al., J. Phys. Chem. B 119(23), 7015-7029 (2015)], we explored large deviations from Stokes-Einstein translational diffusion caused by such a heterogeneity. The current article is set to explore the effect of soft and stiff solvent environments (i.e., structure) on OH-bond rotations in the case of water and small alcohols in 1-ethyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide (Im1,2 (+)NTf2 (-)). Is solute rotational dynamics heterogeneous? If so, are solute rotations and translations coupled in the sense that stiff and soft solvent environments hinder or speed up both types of dynamics? For the systems studied here, there appears to be a clear connection between translations, rotations, and stiff/soft solvent environments. We also discuss interesting asymmetries of the correlation between solutes with anions and cations.

Communication: Nanoscale structure of tetradecyltrihexylphosphonium based ionic liquids.

In a recent communication [J. J. Hettige et al., J. Chem. Phys. 140, 111102 (2014)], we investigated the anomalous temperature dependence of the X-ray first sharp diffraction peak (or prepeak) in the tetradecyltrihexylphosphonium bis(trifluoromethylsulfonyl)-amide ionic liquid. Contrary to what was expected and often observed, the first sharp diffraction peak in this system was shown to increase in intensity with increasing temperature. This implies higher intermediate-range periodicity at a higher temperature. Is this counter-intuitive behavior specific to the combination of cation and anion? The current work analyzes the structural behavior of the same cation coupled with six different anions ranging from the small and spherically symmetric Cl(-) to the more structurally complex and charge-diffuse NTf2 (-). In all cases, the same temperature behavior trend for the prepeak is observed independent of anionic nature. We will show that the intensity increase in the prepeak region is associated with the structural behavior of charged liquid subcomponents. Instead, upon a temperature increase, the apolar subcomponents contribute to what would be an expected decrease of prepeak intensity.

The Role of Balloon Aortic Valvuloplasty in Patients With Aortic Valve Stenosis and Society of Thoracic Surgeons Risk of 15% or Higher.

BACKGROUND: Extreme-risk patients (ie, Society of Thoracic Surgeons [STS] risk 15% or higher) with severe aortic valve stenosis may not obtain mortality benefit from aortic valve replacement (AVR). We reviewed our experience with this group of patients to better understand our triage process and outcomes. METHODS: We performed a retrospective review of 97 patients with severe aortic valve stenosis and STS risk of 15% or higher treated from 2008 through 2013. The median patient age was 85 years (minimum, 44; maximum, 97 years), and 47 patients (48.5%) were male. The STS risk of mortality was 19.8% (minimum, 15.1%; maximum, 60.9%). Patients were assigned to treatment groups based on the first aortic valve intervention of balloon aortic valvuloplasty (BAV group, 66 [68%]) or de novo AVR (d-AVR group, 31 [32%]). RESULTS: Patients in the BAV group were sicker, with a reduced ejection fraction (0.35 vs 0.57; p = 0.002) and greater prevalence of urgent/emergency operative status (32% vs 10%; p = 0.004) compared with those in the d-AVR group. After BAV, 33 patients (50%) demonstrated clinical improvement and went on to receive subsequent staged AVR after a median of 64 days (minimum, 3; maximum, 390 days). The mortality rate at 2 years was worse in the BAV group (57.3% ± 6.3%) than in the d-AVR group (29.5% ± 8.3%; p = 0.015), but was similar in patients who received BAV followed by staged AVR and de novo AVR (p = 0.426). CONCLUSIONS: BAV may triage select patients with STS risk 15% or higher who are questionable candidates for AVR. Patients with clinical improvement after BAV experience benefit from staged AVR.

Ionic liquids-Conventional solvent mixtures, structurally different but dynamically similar.

In more than one way pure ionic liquids (ILs) can be seen as mixtures. By definition they are comprised of cationic and anionic components and they also possess dual charge and apolar characteristics. We recently uncovered interesting dynamical behavior [Araque et al., J. Phys. Chem. B 119(23), 7015 (2015)] that can be ascribed to this duality. For small neutral solutes local friction can be high in certain regions and much lower in others. It is only reasonable to ask whether this interesting behavior is unique to ILs or is also common in certain conventional solvent mixtures such as dimethylsufoxide/glycerol for which the viscosity can be tuned to be similar. We make the case that the latter scenario is correct and that whereas viscous conventional solvent mixtures are structurally very different from ILs, dynamically they are not. From the perspective of a solute that is small, both ILs and viscous conventional solvent mixtures display frictionally stiff and soft regions associated with cage and jump diffusive regimes. In the case of ILs these are associated with charge-enhanced and charge-depleted liquid regions, whereas in the case of the conventional solvents by the distinct frictional properties of the two components.

Modern Room Temperature Ionic Liquids, a Simple Guide to Understanding Their Structure and How It May Relate to Dynamics.

Modern room temperature ionic liquids are structurally defined by symmetries on different length scales. Polar-apolar alternation defines their nanoscale structural heterogeneity, whereas positive-negative charge alternation defines short length scale order. Much progress has been made in the past few years as it pertains to the theoretical interpretation of X-ray scattering experiments for these liquids. Our group has contributed to the development of theoretical interpretation guidelines for the analysis of their structure function. Perhaps less well developed is our understanding of how transport and dynamics in general couple to the very unique structure of ionic liquids which are often dynamically and structurally heterogeneous. This article attempts to present our most current understanding of ionic liquid structure in general and its coupling to transport and dynamics in minimally technical terms for the benefit of the broadest audience.

Effects of Fabrication on Early Patency and Regeneration of Small Intestinal Submucosa Vascular Grafts.

Small intestinal submucosa grafts for vascular regeneration have produced variable patency (0-100%) that has been concurrent with variability in fabrication techniques. We hypothesized that 1) preservation (P) or removal (R) of the stratum compactum layer of the intestine and 2) a dehydrated (D) or hydrated (H) state of the graft, affect early patency and tissue regeneration. We combined both parameters through a 2(2) factorial experimental design into four groups (PD, RD, PH, RH), and compared them in an in vivo early response predictive model (swine, ID 4.5 mm, 7d, n = 4). Patency, thrombogenicity, vascularization, fibroblast infiltration, macrophage polarization profile, endothelialization, and biaxial mechanics were assessed. PD grafts remained patent (4/4) but had scarce vascularization and fibroblast infiltration. RD and RH had extensive vascularization and fibroblast infiltration, however, RD had sustained patency (4/4) and the highest number of regeneration-associated phenotype macrophages (M2), whereas RH had lower patency (3/4) and less M2 macrophages. PH had a modest cellular infiltration, but the lowest patency (2/4) and a dominant adverse macrophage phenotype. Elasticity of R grafts evolved toward that of native carotids (particularly RD), while P grafts kept their initial stiffness. We concluded that fabrication parameters drastically affected early patency and regeneration, with RD providing the best results.

On-pump coronary artery bypass graft operation: Is one crossclamp application better than two?

OBJECTIVES: Several factors may increase the risk of stroke during coronary artery bypass grafting. These include age and atherosclerosis, which are not modifiable, and aortic manipulation, which may be modifiable. This study reports our experience with variable degrees of aortic manipulation (ie, single vs double [partial occlusion] aortic crossclamp techniques) and its influence on rate of operative stroke. METHODS: We performed a retrospective review of 8497 patients treated with isolated on-pump coronary artery bypass grafting from 1993 to 2010. Demographics included an age of 66.8 ± 10.3 years and male sex in 6548 patients (77.1%). Operative technique used the single aortic crossclamp in 2051 patients (24.1%) and the partial aortic crossclamp in 6446 patients (75.9%). To adjust for differences in baseline patient characteristics, 2 propensity-matched cohorts of 1333 patients each were created using Society of Thoracic Surgeons risk calculator variables. RESULTS: In the unmatched cohorts, stroke occurred in 25 patients (1.2%) in the single aortic crossclamp cohort and in 98 patients (1.5%) in the partial aortic crossclamp cohort (P = .320). Logistic regression analysis demonstrated no significant relationship between stroke and aortic occlusion clamp technique (single clamp odds ratio, 0.80; 95% confidence interval, 0.51-1.24; P = .321). In the matched cohorts, stroke occurred in 16 patients (1.2%) in both the single and partial occlusion clamp cohorts (P = 1.00). CONCLUSIONS: Given the methods and limitations of the data analysis, the single and partial aortic crossclamp techniques result in similar rates of stroke during on-pump coronary artery bypass grafting.

How Is Diffusion of Neutral and Charged Tracers Related to the Structure and Dynamics of a Room-Temperature Ionic Liquid? Large Deviations from Stokes-Einstein Behavior Explained.

The deviations from Stokes-Einstein hydrodynamics of small solutes are more pronounced in ionic liquids than in conventional solvents (J. Phys. Chem. B 2013 117 (39), 11697). Small neutral solutes diffuse much faster than expected, whereas small charged solutes diffuse much slower. This article attempts to establish a link between the local friction experienced by tracer solutes and the polar/apolar structure of ionic liquids. We find that small neutral solutes probe locally "stiff" (mostly charged, high electrostriction) regions and locally "soft" (mostly apolar, low electrostriction) regions. These regions of high and low friction are associated with cage and jump regimes. Enhanced neutral tracer mobility in the low friction regions associated with the cationic apolar component has an important bearing on the large positive deviations from Stokes-Einstein behavior. In contrast, diminished charged tracer mobility involves long caging dynamics separated by jump events often triggered by the loss and recovery of counterions.

Molecular dynamics of equilibrium and pressure-driven transport properties of water through LTA-type zeolites.

We consider an atomistic model to investigate the flux of water through thin Linde type A (LTA) zeolite membranes with differing surface chemistries. Using molecular dynamics, we have studied the flow of water under hydrostatic pressure through a fully hydrated LTA zeolite film (~2.5 nm thick) capped with hydrophilic and hydrophobic moieties. Pressure drops in the 50-400 MPa range were applied across the membrane, and the flux of water was monitored for at least 15 ns of simulation time. For hydrophilic membranes, water molecules adsorb at the zeolite surface, creating a highly structured fluid layer. For hydrophobic membranes, a depletion of water molecules occurs near the water/zeolite interface. For both types of membranes, the water structure is independent of the pressure drop established in the system and the flux through the membranes is lower than that observed for the bulk zeolitic material; the latter allows an estimation of surface barrier effects to pressure-driven water transport. Mechanistically, it is observed that (i) bottlenecks form at the windows of the zeolite structure, preventing the free flow of water through the porous membrane, (ii) water molecules do not move through a cage in a single-file fashion but rather exhibit a broad range of residence times and pronounced mixing, and (iii) a periodic buildup of a pressure difference between inlet and outlet cages takes place which leads to the preferential flow of water molecules toward the low-pressure cages.

A theoretical and simulation study of the self-assembly of a binary blend of diblock copolymers.

Pure diblock copolymer melts exhibit a narrow range of conditions at which bicontinuous and cocontinuous phases are stable; such conditions and the morphology of such phases can be tuned by the use of additives. In this work, we have studied a bidisperse system of diblock copolymers using theory and simulation. In particular, we elucidated how a short, lamellar-forming diblock copolymer modifies the phase behavior of a longer, cylinder-forming diblock copolymer. In a narrow range of intermediate compositions, self-consistent field theory predicts the formation of a gyroid phase although particle-based simulations show that three phases compete: the gyroid phase, a disordered cocontinuous phase, and the cylinder phase, all having free energies within error bars of each other. Former experimental studies of a similar system have yielded an unidentified, partially irregular bicontinuous phase, and our simulations suggest that at such conditions the formation of a partially transformed network phase is indeed plausible. Close examination of the spatial distribution of chains reveals that packing frustration (manifested by chain stretching and low density spots) occurs in the majority-block domains of the three competing phases simulated. In all cases, a double interface around the minority-block domains is also detected with the outer one formed by the short chains, and the inner one formed by the longer chains.

Lattice model of oligonucleotide hybridization in solution. I. Model and thermodynamics.

A coarse-grained lattice model of DNA oligonucleotides is proposed to investigate the general mechanisms by which single-stranded oligonucleotides hybridize to their complementary strands in solution. The model, based on a high-coordination cubic lattice, is simple enough to allow the direct simulation of DNA solutions, yet capturing how the fundamental thermodynamic processes are microscopically encoded in the nucleobase sequences. Physically relevant interactions are considered explicitly, such as interchain excluded volume, anisotropic base-pairing and base-stacking, and single-stranded bending rigidity. The model is studied in detail by a specially adapted Monte Carlo simulation method, based on parallel tempering and biased trials, which is designed to overcome the entropic and enthalpic barriers associated with the sampling of hybridization events of multiple single-stranded chains in solution. This methodology addresses both the configurational complexity of bringing together two complementary strands in a favorable orientation (entropic barrier) and the energetic penalty of breaking apart multiple associated bases in a double-stranded state (enthalpic barrier). For strands with sequences restricted to nonstaggering association and homogeneous pairing and stacking energies, base-pairing is found to dominate the hybridization over the translational and conformational entropy. For strands with sequence-dependent pairing corresponding to that of DNA, the complex dependence of the model's thermal stability on concentration, sequence, and degree of complementarity is shown to be qualitatively and quantitatively consistent both with experiment and with the predictions of statistical mechanical models.

Transition path sampling and forward flux sampling. Applications to biological systems.

The last decade has seen a rapid growth in the number of simulation methods and applications dealing with the sampling of transition pathways of rare nanoscale events. Such studies are crucial, for example, for understanding the mechanism and kinetics of conformational transitions and enzymatic events associated with the function of biomolecules. In this review, a broad account of transition path sampling approaches is provided, starting from the general concepts, progressing to the specific principles that underlie some of the most important methods, and eventually singling out the so-called forward flux sampling method for a more detailed description. This is done because forward flux sampling, despite its appealing simplicity and potential efficiency, has thus far received limited attention from practitioners. While path sampling methods have a widespread application to many types of rare transitional events, here only recent applications involving biomolecules are reviewed, including isomerization, protein folding, and enzyme catalysis.