Aspiring to Unintended Consequences of Natural Language Processing: A Review of Recent Developments in Clinical and Consumer-Generated Text Processing.

OBJECTIVES: This paper reviews work over the past two years in Natural Language Processing (NLP) applied to clinical and consumer-generated texts. METHODS: We included any application or methodological publication that leverages text to facilitate healthcare and address the health-related needs of consumers and populations. RESULTS: Many important developments in clinical text processing, both foundational and task-oriented, were addressed in community- wide evaluations and discussed in corresponding special issues that are referenced in this review. These focused issues and in-depth reviews of several other active research areas, such as pharmacovigilance and summarization, allowed us to discuss in greater depth disease modeling and predictive analytics using clinical texts, and text analysis in social media for healthcare quality assessment, trends towards online interventions based on rapid analysis of health-related posts, and consumer health question answering, among other issues. CONCLUSIONS: Our analysis shows that although clinical NLP continues to advance towards practical applications and more NLP methods are used in large-scale live health information applications, more needs to be done to make NLP use in clinical applications a routine widespread reality. Progress in clinical NLP is mirrored by developments in social media text analysis: the research is moving from capturing trends to addressing individual health-related posts, thus showing potential to become a tool for precision medicine and a valuable addition to the standard healthcare quality evaluation tools.

UMLS-based automatic image indexing.

To date, most accurate image retrieval techniques rely on textual descriptions of images. Our goal is to automatically generate indexing terms for an image extracted from a biomedical article by identifying Unified Medical Language System (UMLS) concepts in image caption and its discussion in the text. In a pilot evaluation of the suggested image indexing method by five physicians, a third of the automatically identified index terms were found suitable for indexing.

Peptide chemical ligation inside living cells: in vivo generation of a circular protein domain.

Here we describe the first example of a peptide chemical ligation reaction performed inside a living cell. A cell-based native chemical ligation approach was developed and used to generate a circular version of the N-terminal Src homology 3 (SH3) domain from the murine c-Crk adapter protein inside Escherichia coli cells. The in vivo cyclization reaction was extremely efficient and the resulting circular protein domain was fully biologically active and able to adopt the native SH3 folded structure. This work represents an important step towards the in vivo generation of small backbone cyclic peptides for use in basic biological research.

Simulated and NMR-derived backbone dynamics of a protein with significant flexibility: a comparison of spectral densities for the betaARK1 PH domain.

A 7.6 ns molecular dynamics trajectory of the betaARK1 PH domain in explicit water with appropriate ions was calculated at 300 K. Spectral densities at omega = 0, omega(N), and 0.87omega(H) and the model-free parameters were evaluated from the experimental as well as the simulated data, taking the anisotropic overall motion of the protein into account. Experimental and simulated spectral densities are in reasonable general agreement for NH bond vectors, where the corresponding motions have converged within the simulation time. A sufficient sampling of the motions for NH bonds within flexible parts of the protein requires a longer simulation time. The simulated spectral densities J(0) and J(omega(N)) are, on average, 4.5% and 16% lower than the experimental data; the corresponding numbers for the core residues are about 6%; the high-frequency spectral densities J(0.87omega(H)) are lower by, on average, 16% (21% for the core). The simulated order parameters, S(2), are also lower, although the overall disagreement between the simulation and experiment is less pronounced: 1% for all residues and 6% for the core. The observed systematic decrease of simulated spectral density and the order parameters compared to the experimental data can be partially attributed to the ultrafast librational motion of the NH bonds with respect to their peptide plane, which was analyzed in detail. This systematic difference is most pronounced for J(0.87omega(H)), which appears to be most sensitive to the slow, subnanosecond time scale of internal motion, whereas J(0) and J(omega(N)) are dominated by the overall rotational tumbling of the protein. Similar discrepancies are observed between the experimentally measured (15)N relaxation parameters (R(1), R(2), NOE) and their values calculated from the simulated spectral densities. The analysis of spectral densities provides additional information regarding the comparison of the simulated and experimental data, not available from the model-free analysis.

Nuclear magnetic resonance relaxation in determination of residue-specific 15N chemical shift tensors in proteins in solution: protein dynamics, structure, and applications of transverse relaxation optimized spectroscopy.

We developed several approaches to direct determination of the 15N CSA from relaxation measurements in uniformly 15N-labeled proteins in solution. These methods are based on multiple-field measurements and could be extended to other nuclei in proteins and other molecules. Combined with the isotropic chemical shift measurements, this provides an experimental approach to full characterization of chemical shift tensors in proteins in their native milieu, which is likely to provide valuable information on the nature of chemical shifts and their relation to protein structure. Knowledge of 15N CSA is essential for an accurate characterization of protein dynamics from relaxation measurements.

Rescuing a destabilized protein fold through backbone cyclization.

We describe the physicochemical characterization of various circular and linear forms of the approximately 60 residue N-terminal Src homology 3 (SH3) domain from the murine c-Crk adapter protein. Structural, dynamic, thermodynamic, kinetic and biochemical studies reveal that backbone circularization does not prevent the adoption of the natural folded structure in any of the circular proteins. Both the folding and unfolding rate of the protein increased slightly upon circularization. Circularization did not lead to a significant thermodynamic stabilization of the full-length protein, suggesting that destabilizing enthalpic effects (e.g. strain) negate the expected favorable entropic contribution to overall stability. In contrast, we find circularization results in a dramatic stabilization of a truncated version of the SH3 domain lacking a key glutamate residue. The ability to rescue the destabilized mutant indicates that circularization may be a useful tool in protein engineering programs geared towards generating minimized proteins.

Determination of the rotational diffusion tensor of macromolecules in solution from nmr relaxation data with a combination of exact and approximate methods--application to the determination of interdomain orientation in multidomain proteins.

In this paper we present a method for determining the rotational diffusion tensor from NMR relaxation data using a combination of approximate and exact methods. The approximate method, which is computationally less intensive, computes values of the principal components of the diffusion tensor and estimates the Euler angles, which relate the principal axis frame of the diffusion tensor to the molecular frame. The approximate values of the principal components are then used as starting points for an exact calculation by a downhill simplex search for the principal components of the tensor over a grid of the space of Euler angles relating the diffusion tensor frame to the molecular frame. The search space of Euler angles is restricted using the tensor orientations calculated using the approximate method. The utility of this approach is demonstrated using both simulated and experimental relaxation data. A quality factor that determines the extent of the agreement between the measured and predicted relaxation data is provided. This approach is then used to estimate the relative orientation of SH3 and SH2 domains in the SH(32) dual-domain construct of Abelson kinase complexed with a consolidated ligand.

The virtual NMR spectrometer: a computer program for efficient simulation of NMR experiments involving pulsed field gradients.

This paper presents a software program, the Virtual NMR Spectrometer, for computer simulation of multichannel, multidimensional NMR experiments on user-defined spin systems. The program is capable of reproducing most features of the modern NMR experiment, including homo- and heteronuclear pulse sequences, phase cycling, pulsed field gradients, and shaped pulses. Two different approaches are implemented to simulate the effect of pulsed field gradients on coherence selection, an explicit calculation of all coherence transfer pathways, and an effective approximate method using integration over multiple positions in the sample. The applications of the Virtual NMR Spectrometer are illustrated using homonuclear COSY and DQF COSY experiments with gradient selection, heteronuclear HSQC, and TROSY. The program uses an intuitive graphical user interface, which resembles the appearance and operation of a real spectrometer. A translator is used to allow the user to design pulse sequences with the same programming language used in the actual experiment on a real spectrometer. The Virtual NMR Spectrometer is designed as a useful tool for developing new NMR experiments and for tuning and adjusting the experimental setup for existing ones prior to running costly NMR experiments, in order to reduce the setup time on a real spectrometer. It will also be a useful aid for learning the general principles of magnetic resonance and contemporary innovations in NMR pulse sequence design.

Direct determination of changes of interdomain orientation on ligation: use of the orientational dependence of 15N NMR relaxation in Abl SH(32).

The relative orientation and motions of domains within many proteins are key to the control of multivalent recognition, or the assembly of protein-based cellular machines. Current methods of structure determination have limited applicability to macromolecular assemblies, characterized by weak interactions between the constituents. Crystal structures of such complexes might be biased by packing forces comparable to the interdomain interactions, while the precision and accuracy of the conventional NMR structural approaches are necessarily limited by the restricted number of NOE contacts and by interdomain flexibility rendering the available NOE information uninterpretable. NMR relaxation studies are capable of providing "long-range" structural information on macromolecules in their native milieu. Here we determine directly the change in domain orientation between unligated and dual ligated subdomains of the SH(32) segment of Abelson kinase in solution, using the orientational dependence of nuclear spin relaxation. These results demonstrate that the change in domain orientation between unligated and ligated forms can be measured directly in solution.

A comparative study of the backbone dynamics of two closely related lipid binding proteins: bovine heart fatty acid binding protein and porcine ileal lipid binding protein.

The backbone dynamics of bovine heart fatty acid binding protein (H-FABP) and porcine ileal lipid binding protein (ILBP) were studied by 15N NMR relaxation (T1 and T2) and steady state heteronuclear 15N[1H] NOE measurements. The microdynamic parameters characterizing the backbone mobility were determined using the 'model-free' approach. For H-FABP, the non-terminal backbone amide groups display a rather compact protein structure of low flexibility. In contrast, for ILBP an increased number of backbone amide groups display unusually high internal mobility. Furthermore, the data indicate a higher degree of conformational exchange processes in the microsec-msec time range for ILBP compared to H-FABP. These results suggest significant differences in the conformational stability for these two structurally highly homologous members of the fatty acid binding protein family.

Impact of Cl- and Na+ ions on simulated structure and dynamics of betaARK1 PH domain.

A nonzero net charge of proteins at pH 7 is usually compensated by the addition of charge-balancing counter ions during molecular dynamics simulation, which reduces electrostatic interactions. For highly charged proteins, like the betaARK1 PH domain used here, it seems reasonable to also add explicit salt ions. To assess the impact of explicit salt ions, two molecular dynamics simulations of solvated betaARK1 PH domain have been carried out with different numbers of Cl- and Na+ ions, based on the Cornell et al. force field and the Ewald summation, which was used in the treatment of long-range electrostatic interactions. Initial positions of ions were obtained from the AMBER CION program. Increasing the number of ions alters the average structure in loop regions, as well as the fluctuation amplitudes of dihedral angles. We found unnaturally strong interactions between side chains in the absence of salt ions. The presence of salt ions reduces these electrostatic interactions. The time needed for the equilibration of the ionic environment around the protein, after initial placement of ions close to oppositely charged side chains, is in the nanosecond time range, which can be shortened by using a higher ionic strength. Our results also suggest selecting those methods that do not place the ions initially close to the protein surface.

Solution structure of the proapoptotic molecule BID: a structural basis for apoptotic agonists and antagonists.

Members of the BCL2 family of proteins are key regulators of programmed cell death, acting either as apoptotic agonists or antagonists. Here we describe the solution structure of BID, presenting the structure of a proapoptotic BCL2 family member. An analysis of sequence/structure of BCL2 family members allows us to define a structural superfamily, which has implications for general mechanisms for regulating proapoptotic activity. It appears two criteria must be met for proapoptotic function within the BCL2 family: targeting of molecules to intracellular membranes, and exposure of the BH3 death domain. BID's activity is regulated by a Caspase 8-mediated cleavage event, exposing the BH3 domain and significantly changing the surface charge and hydrophobicity, resulting in a change of cellular localization.

The effect of noncollinearity of 15N-1H dipolar and 15N CSA tensors and rotational anisotropy on 15N relaxation, CSA/dipolar cross correlation, and TROSY.

Current approaches to 15N relaxation in proteins assume that the 15N-1H dipolar and 15N CSA tensors are collinear. We show theoretically that, when there is significant anisotropy of molecular rotation, different orientations of the two tensors, experimentally observed in proteins, nucleic acids, and small peptides, will result in differences in site-specific correlation functions and spectral densities. The standard treatments of the rates of longitudinal and transverse relaxation of amide 15N nuclei, of the 15N CSA/15N-1H dipolar cross correlation, and of the TROSY experiment are extended to account for the effect of noncollinearity of the 15N-1H dipolar and 15N CSA (chemical shift anisotropy) tensors. This effect, proportional to the degree of anisotropy of the overall motion, (D parallel/D perpendicular - 1), is sensitive to the relative orientation of the two tensors and to the orientation of the peptide plane with respect to the diffusion coordinate frame. The effect is negligible at small degrees of anisotropy, but is predicted to become significant for D parallel/D perpendicular > or = 1.5, and at high magnetic fields. The effect of noncollinearity of 15N CSA and 15N-1H dipolar interaction is sensitive to both gross (hydrodynamic) properties and atomic-level details of protein structure. Incorporation of this effect into relaxation data analysis is likely to improve both precision and accuracy of the derived characteristics of protein dynamics, especially at high magnetic fields and for molecules with a high degree of anisotropy of the overall motion. The effect will also make TROSY efficiency dependent on local orientation in moderately anisotropic systems.

Solution structure and dynamics of the bioactive retroviral M domain from Rous sarcoma virus.

A biologically active construct of the retroviral M domain from the avian Rous sarcoma virus is defined and its solution structure described. This M domain is fully active in budding and infectivity without myristylation. In spite of a sequence homology level that suggests no relationship among M domains and the family of matrix proteins in mammalian retroviruses, the conserved structural elements of a central core allow an M domain sequence motif to be described for all retroviruses. The surface of the M domain has a highly clustered positive patch comprised of sequentially distant residues. An analysis of the backbone dynamics, incorporating rotational anisotropy, is used to estimate the thermodynamics of proposed domain oligomerization.

The solution structure and dynamics of the pleckstrin homology domain of G protein-coupled receptor kinase 2 (beta-adrenergic receptor kinase 1). A binding partner of Gbetagamma subunits.

The solution structure of an extended pleckstrin homology (PH) domain from the beta-adrenergic receptor kinase is obtained by high resolution NMR. The structure establishes that the beta-adrenergic receptor kinase extended PH domain has the same fold and topology as other PH domains, and there are several unique features, most notably an extended C-terminal alpha-helix that behaves as a molten helix, and a surface charge polarity that is extensively modified by positive residues in the extended alpha-helix and the C terminus. These observations complement biochemical evidence that the C-terminal portion of this PH domain participates in protein-protein interactions with Gbetagamma subunits. This suggests that the C-terminal segment of the PH domain may function to mediate protein-protein interactions with the targets of PH domains.

The main-chain dynamics of the dynamin pleckstrin homology (PH) domain in solution: analysis of 15N relaxation with monomer/dimer equilibration.

The backbone dynamics of the pleckstrin homology (PH) domain from dynamin were studied by 15N NMR relaxation (R1 and R2) and steady state heteronuclear 15N [1H] nuclear Overhauser effect measurements at 500 and 600 MHz, at protein concentrations of 1.7 mM and 300 microM, and by molecular dynamics (MD) simulations. The analysis was performed using the model-free approach. The method was extended in order to account for observed partial (equilibrium) dimerization of the protein at NMR concentrations. A model is developed that takes into account both rapid monomer-dimer exchange and anisotropy of the over-all rotation of the dimer. The data show complex dynamics of the dynamin PH domain. Internal motions in elements of the secondary structure are restricted, as inferred from the high value of the order parameter (S2 approximately 0.9) and from the local correlation time < 100 ps. Of the four extended loop regions that are disordered in the NMR-derived solution structure of the protein, loops beta 1/beta 2 and beta 5/beta 6 are involved in a large-amplitude (S2 down to 0.2 to 0.3) subnanosecond to nanosecond time-scale motion. Reorientation of the loops beta 3/beta 4 and beta 6/beta 7, in contrast, is restricted, characterized by the values of order parameter S2 approximately 0.9 more typical of the protein core. These loops, however, are involved in much slower processes of motion resulting in a conformational exchange on a microsecond to submillisecond time scale. The motions of the terminal regions (residues 1 to 10, 122 to 125) are practically unrestricted (S2 down to 0.05, characteristic times in nanosecond time scale), suggesting that these parts of the sequence do not participate in the protein fold. The analysis shows a larger sensitivity of the 15N relaxation data to protein microdynamic parameters (S2, tau loc) when protein molecular mass (tau c) increases. The use of negative values of the steady state 15N[1H] NOEs as an indicator of the residues not belonging to the folded structure is suggested. The amplitudes of local motion observed in the MD simulation are in a good-agreement with the NMR data for the amide NH groups located in the protein core.

Identification of the binding site for acidic phospholipids on the pH domain of dynamin: implications for stimulation of GTPase activity.

It has recently been suggested that pleckstrin homology (PH) domains bind specifically to phospholipids, with phosphatidylinositol-4,5-bisphosphate (PtdIns(4,5)P2) being most strongly bound. This observation suggests that PH domains may be responsible for membrane association of proteins in which they occur. Further, this membrane association may be regulated by enzymes that modify lipid head groups to which PH domains may bind. We have studied the binding of phospholipids to the PH domain of human dynamin, a 100 kDa GTPase that is involved in the initial stages of endocytosis. We describe a rapid method for screening PH domain/ligand interactions that gives precise binding constants. We confirm that PtdIns(4,5)P2 can bind to dynamin PH domain, although not in an aggregated state. Using NMR spectroscopy, we have mapped a specific site on the surface of dynamin PH domain of which binding of gIns(1,4,5)P3 (the head-group skeleton of PtdIns(4,5)P2) occurs. The relative affinity of acidic phospholipids for dynamin PH domain correlates with their ability to activate the GTPase of dynamin. We propose, therefore, that the interaction of these phospholipids with dynamin is likely to occur via the PH domain. Given the fact that PH domains are often found in proteins associated with GTPase activity, or in guanine nucleotide exchange factors, we suggest that one role of PH domains may be to couple phosphatidylinositol signalling to GTP hydrolysis.

Solution structure of pleckstrin homology domain of dynamin by heteronuclear NMR spectroscopy.

The pleckstrin homology (PH) domain is a recognition motif thought to be involved in signal-transduction pathways controlled by a variety of cytoplasmic proteins. Assignments of nearly all 1H, 13C, and 15N resonances of the PH domain from dynamin have been obtained from homonuclear and heteronuclear NMR experiments. The secondary structure has been elucidated from the pattern of nuclear Overhauser enhancements, from 13C chemical shift deviations, and from observation of slowly exchanging amide hydrogens. The secondary structure contains one alpha-helix and eight beta-strands, seven of which are arranged in two contiguous, antiparallel beta-sheets. The structure is monomeric, in contrast to the well-defined intimate dimerization of the crystal structure of this molecule. Residues possibly involved in ligand binding are in apparently flexible loops. Steady-state 15N(1H) nuclear Overhauser effect measurements indicate unequivocally the boundaries of this PH domain, and the structured portion of the domain appears to be more extended to the C terminus than previously suggested for other PH domains.

Determination of the backbone mobility of ribonuclease T1 and its 2'GMP complex using molecular dynamics simulations and NMR relaxation data.

The results of 1-nanosecond molecular dynamics simulations of the enzyme ribonuclease T1 and its 2'GMP complex in water are presented. A classification of the angular reorientations of the backbone amide groups is achieved via a transformation of NH-vector trajectories into several coordinate frames, thus unravelling contributions of NH-bond librations and backbone dihedral angle fluctuations. The former turned out to be similar for all amides, as characterized by correlation times of librational motions in a subpicosecond scale, angular amplitudes of about 10-12 degrees for out-of-peptide-plane displacements of the NH-bond and 3-5 degrees for the in-plane displacements, whereas the contributions of much slower backbone dihedral angle fluctuations strongly depend on the secondary structure. Correlation functions relevant for NMR were obtained and analyzed utilizing the 'model-free' approach (Lipari, G. and Szabo, A. (1982) J. Am. Chem. Soc. 104, 4546-4559, 4559-4570; Clore et al., (1990) J. Am. Chem. Soc. 112, 4989-4991). The dependence of the amplitude of local motion on the residue location in the backbone is in good agreement with the results of NMR relaxation measurements and X-ray data. The protein dynamics is characterized by a highly restricted local motion of those parts of the backbone with defined secondary structure as well as by a high flexibility in loop regions. The comparison of results derived from different periods of the trajectory (of 50 ps and 1 ns duration, 1000 points sampled) reveals a dependence of the observed dynamic picture on the characteristic time scale of the experimental method used. Comparison of the MD data for the free and liganded enzyme clearly indicates a restriction of the mobility within certain regions of the backbone upon inhibitor binding.