Nonlinear reconfiguration of network edges, topology and information content during an artificial learning task.

Here, we combine network neuroscience and machine learning to reveal connections between the brain's network structure and the emerging network structure of an artificial neural network. Specifically, we train a shallow, feedforward neural network to classify hand-written digits and then used a combination of systems neuroscience and information-theoretic tools to perform 'virtual brain analytics' on the resultant edge weights and activity patterns of each node. We identify three distinct phases of network reconfiguration across learning, each of which are characterized by unique topological and information-theoretic signatures. Each phase involves aligning the connections of the neural network with patterns of information contained in the input dataset or preceding layers (as relevant). We also observe a process of low-dimensional category separation in the network as a function of learning. Our results offer a systems-level perspective of how artificial neural networks function-in terms of multi-stage reorganization of edge weights and activity patterns to effectively exploit the information content of input data during edge-weight training-while simultaneously enriching our understanding of the methods used by systems neuroscience.

Influence of coracoid anatomy on the location of glenoid rim defects in anterior shoulder instability: 3D CT-scan evaluation of 51 patients.

PURPOSE: Glenoid bony lesions play a role in approximately half of anterior shoulder instability cases. The purpose of this study is to see if the anatomy of the coracoid affects the location of glenoid rim defects. We hypothesized that a prominent coracoid (lower and lateral) would be more likely to cause an anterior-inferior glenoid lesion, and a less prominent coracoid more prone to cause an anterior lesion. The null hypothesis being the absence of correlation. METHODS: Fifty-one shoulder CT-scans from a prospective database, with 3D reconstruction, were analyzed. The position of glenoid lesions was identified using the validated clock method, identifying the beginning and end time. The size of bony glenoid defects was calculated using the validated glenoid ratio method. The position of the coracoid tip was measured in three orthogonal planes. RESULTS: Analysis included 25 right shoulders and 26 left shoulders in seven females and 41 males. The vertical position of the coracoid tip relative to the top of the glenoid was highly correlated to the location of the glenoid defect on the profile view (r = -0.625; 95% CI 0.423-0.768; p = 0.001). Thus, higher coracoids were associated with anterior lesions, while lower coracoids were associated with anterior-inferior lesions. A more laterally prominent coracoid was also correlated with anterior-inferior lesions (r = 0.433; 95% CI 0.179-0.633; p = 0.002). CONCLUSION: This study shows that coracoid anatomy affects the location of bony Bankart defects in anterior shoulder instability. Lower and laterally prominent coracoids are associated with anterior-inferior lesions. This variation in anatomy should be considered during pre-op planning for surgeries involving bone graft. LEVEL OF EVIDENCE: Level 4 basic science.

Topological Properties of Neuromorphic Nanowire Networks.

Graph theory has been extensively applied to the topological mapping of complex networks, ranging from social networks to biological systems. Graph theory has increasingly been applied to neuroscience as a method to explore the fundamental structural and functional properties of human neural networks. Here, we apply graph theory to a model of a novel neuromorphic system constructed from self-assembled nanowires, whose structure and function may mimic that of human neural networks. Simulations of neuromorphic nanowire networks allow us to directly examine their topology at the individual nanowire-node scale. This type of investigation is currently extremely difficult experimentally. We then apply network cartographic approaches to compare neuromorphic nanowire networks with: random networks (including an untrained artificial neural network); grid-like networks and the structural network of C. elegans. Our results demonstrate that neuromorphic nanowire networks exhibit a small-world architecture similar to the biological system of C. elegans, and significantly different from random and grid-like networks. Furthermore, neuromorphic nanowire networks appear more segregated and modular than random, grid-like and simple biological networks and more clustered than artificial neural networks. Given the inextricable link between structure and function in neural networks, these results may have important implications for mimicking cognitive functions in neuromorphic nanowire networks.

Transitions in information processing dynamics at the whole-brain network level are driven by alterations in neural gain.

A key component of the flexibility and complexity of the brain is its ability to dynamically adapt its functional network structure between integrated and segregated brain states depending on the demands of different cognitive tasks. Integrated states are prevalent when performing tasks of high complexity, such as maintaining items in working memory, consistent with models of a global workspace architecture. Recent work has suggested that the balance between integration and segregation is under the control of ascending neuromodulatory systems, such as the noradrenergic system, via changes in neural gain (in terms of the amplification and non-linearity in stimulus-response transfer function of brain regions). In a previous large-scale nonlinear oscillator model of neuronal network dynamics, we showed that manipulating neural gain parameters led to a 'critical' transition in phase synchrony that was associated with a shift from segregated to integrated topology, thus confirming our original prediction. In this study, we advance these results by demonstrating that the gain-mediated phase transition is characterized by a shift in the underlying dynamics of neural information processing. Specifically, the dynamics of the subcritical (segregated) regime are dominated by information storage, whereas the supercritical (integrated) regime is associated with increased information transfer (measured via transfer entropy). Operating near to the critical regime with respect to modulating neural gain parameters would thus appear to provide computational advantages, offering flexibility in the information processing that can be performed with only subtle changes in gain control. Our results thus link studies of whole-brain network topology and the ascending arousal system with information processing dynamics, and suggest that the constraints imposed by the ascending arousal system constrain low-dimensional modes of information processing within the brain.

Peak alignment of gas chromatography-mass spectrometry data with deep learning.

We present ChromAlignNet, a deep learning model for alignment of peaks in Gas Chromatography-Mass Spectrometry (GC-MS) data. In GC-MS data, a compound's retention time (RT) may not stay fixed across multiple chromatograms. To use GC-MS data for biomarker discovery requires alignment of identical analyte's RT from different samples. Current methods of alignment are all based on a set of formal, mathematical rules. We present a solution to GC-MS alignment using deep learning neural networks, which are more adept at complex, fuzzy data sets. We tested our model on several GC-MS data sets of various complexities and analysed the alignment results quantitatively. We show the model has very good performance (AUC ∼ 1 for simple data sets and AUC ∼ 0.85 for very complex data sets). Further, our model easily outperforms existing algorithms on complex data sets. Compared with existing methods, ChromAlignNet is very easy to use as it requires no user input of reference chromatograms and parameters. This method can easily be adapted to other similar data such as those from liquid chromatography. The source code is written in Python and available online.

Development of cardiac safety translational tools for QT prolongation and torsade de pointes.

OBJECTIVE: A regulatory science priority at the Food and Drug Administration (FDA) is to promote the development of new innovative tools such as reliable and validated computational (in silico) models. This FDA Critical Path Initiative project involved the development of predictive clinical computational models for decision-support in CDER evaluations of QT/QTc interval prolongation and proarrhythmic potential for non-antiarrhythmic drugs. METHODS: Several classification models were built using predictive technologies of quantitative structure-activity relationship analysis using clinical in-house and public data on induction of QT prolongation and torsade de pointes (TdP) in humans. Specific models were geared toward prediction of high-risk drugs with attention to outcomes from thorough QT studies and TdP risk based on clinical in-house data. Models used were independent of non-clinical data or known molecular mechanisms. The positive predictive performance of the in silico models was validated using cross-validation and independent external validation test sets. RESULTS: Optimal performance was observed with high sensitivity (87%) and high specificity (88%) for predicting QT interval prolongation using in-house data, and 77% sensitivity in predicting drugs withdrawn from the market. Furthermore, the article describes alerting substructural features based on drugs tested in the clinical trials. CONCLUSIONS: The in silico models provide evidence of a structure-based explanation for these cardiac safety endpoints. The models will be made publically available and are under continual prospective external validation testing and updating at CDER using TQT study outcomes.

Highly automated QT measurement techniques in 7 thorough QT studies implemented under ICH E14 guidelines.

Thorough QT (TQT) studies are designed to evaluate potential effect of a novel drug on the ventricular repolarization process of the heart using QTc prolongation as a surrogate marker for torsades de pointes. The current process to measure the QT intervals from the thousands of electrocardiograms is lengthy and expensive. In this study, we propose a validation of a highly automatic-QT interval measurement (HA-QT) method. We applied a HA-QT method to the data from 7 TQT studies. We investigated both the placebo and baseline-adjusted QTc interval prolongation induced by moxifloxacin (positive control drug) at the time of expected peak concentration. The comparative analysis evaluated the time course of moxifloxacin-induced QTc prolongation in one study as well. The absolute HA-QT data were longer than the FDA-approved QTc data. This trend was not different between ECGs from the moxifloxacin and placebo arms: 9.6 ± 24 ms on drug and 9.8 ± 25 ms on placebo. The difference between methods vanished when comparing the placebo-baseline-adjusted QTc prolongation (1.4 ± 2.8 ms, P = 0.4). The differences in precision between the HA-QT and the FDA-approved measurements were not statistically different from zero: 0.1 ± 0.1 ms (P = 0.7). Also, the time course of the moxifloxacin-induced QTc prolongation adjusted for placebo was not statistically different between measurements methods.