Neuroscience Scaffolded by Informatics: A Raging Interdisciplinary Field

Abstract

Following breakthrough achievements in molecular neurosciences, the current decade witnesses a trend toward interdisciplinary and multimodal development. Supplementation of neurosciences with tools from computer science solidifies previous knowledge and sets the ground for new research on “big data” and new hypothesis-free experimental models. In this Special Issue, we set the focus on informatics-supported interdisciplinary neuroscience accomplishments symmetrically combining wet-lab and clinical routines. Video-tracking and automated mitosis detection in vitro, the macromolecular modeling of kinesin motion, and the unsupervised classification of the brain’s macrophage activation status share a common denominator: they are energized by machine and deep learning. Essential clinical neuroscience questions such as the estimated risk of brain aneurysm rupture and the surgical outcome of facial nerve transplantation are addressed in this issue as well. Precise and rapid evaluation of complex clinical data by deep learning and data mining dives deep to reveal symmetrical and asymmetrical features beyond the abilities of human perception or the limits of linear algebraic modeling. This editorial opts to motivate researchers from the wet lab, computer science, and clinical environments to join forces in reshaping scientific platforms, share and converge high-quality data on public platforms, and use informatics to facilitate interdisciplinary information exchange.

1. Introduction

Neuroscience is a multidisciplinary branch interconnected with molecular biology, immunology, computer vision, deep learning, and clinical entities such as endovascular surgery and public health. Brain research and research on the peripheral nervous system extend beyond their classical borders in interdisciplinary and multimodal directions. Novel supplementation of neuroscience methods with tools from computer science solidifies old knowledge with unbiased evidence, opens new horizons, and allows for “big data” administration towards new hypothesis-free experimental models. Computer vision supplies new methods or reinforces old approaches into tools of increased potential to support the needs and challenges of the upcoming decade.

In this Special Issue, we set the focus on informatics-supported interdisciplinary neuroscience accomplishments designed to improve wet-lab routines and promote a faster, high-accuracy bench-to-bedside translation. Video-tracking and automated detection of mitosis in vitro [1], the macromolecular modeling of kinesin motion [2], and the assumption-free functional profiling of brain’s macrophages [3] have as a common denominator a deep learning model. Essential clinical neuroscience questions such as the estimated risk of brain aneurysm rupture [4] and the surgical outcome of facial nerve transplantation [5] require rapid and complex converging data handling, enabled by deep learning methods that perform beyond the limits of linear algebraic modeling.

2. Body

The intracellular transport of macromolecules and vesicles is an essential housekeeping and turnover function, attracting the particular interest of neuroscientists for supporting the communication between the neural soma and distal processes, namely axons and dendrites. The intracellular cargo system extends beyond physiology to include dysfunctions in neurodegenerative disorders with potential pharmacological targeting. Through a dimerization-based mechanochemical coupling, kinesin and dynein bind to microtubules and progress stepwise in a forward or backward motion, supplying distal neuronal compartments with necessary cargoes [6]. Kinesin and dynein, as the primary cell cargo vehicles, essentially sustain the synaptic transport [7,8]; kinesin is the core antegrade axonal transport system with cargo to sustain synaptic function, whereas dynein leads the main retrograde trafficking cell facility. The visualization of kinesin motion is enabled by optical trapping methods using trackable nanosphere reporters [8]. Recent results for a truncated rat kinesin trapped by germanium nanospheres [9] showed that kinesin-1 walks with steps of 4 nm. Xie [2] reproduced the experimental results of Sudhakar et al. [9] in silico by developing a theoretical model of kinesin mechanochemical coupling. The model envelops the hypothesis of a load-dependent, probabilistic ante- and retrograde trajectory and could predict the experimental results on the transport speed of kinesin depending on the cargo load. The in silico model of Xie [2] prompts future challenges and inspires novel horizons, such as studying recently discovered kinesin variants [10,11] that do not obey the classical mechanochemical models [12].

Beyond the field of bioengineering, informatics has transformed the landscape of quantitative microscopy. The technical evolution of time-lapse video microscopy generated big data and created the urge for automated data evaluation. The hosted publication by Hadjidemetriou et al. [1] introduces and tests a pipeline for the automated detection and segmentation of mitosis using phase contrast (PC) microscopy video time-lapses in cell cultures. PC is a light microscopy technique using visible light polarization to enhance cell contours. PC imaging is a low-energy and low-cost method but does not allow for molecular reporting or tagged cell tracking. Hadjidemetriou et al. introduce a mitosis detection and segmentation pipeline that boosts the diagnostic value of PC imaging without additional lab costs. Tedious image reconstruction and restoration is the cornerstone for a robust detection and segmentation step. Noise and intensity non-uniformities are compensated for in time and space using an iterative design with quality thresholding to avoid eternal loops. Cells are modeled as centroids with a pixel-based trackable trajectory over time and space, reaching a final sensitivity and precision of 0.85/0.95. The pipeline of Hadjidemetriou et al. [1] is a qualitative example of pixel-based tracking in video lapses. Whereas the model is robust against false positives, false negatives near the temporal and spatial lapse borders could be improved in future research. Novel convolutional neuronal network (CNN) models detecting within- and cross-image long-range dependencies with a semantic association in video time-lapses might increase efficacy and reduce false negatives. Such applications are being tested on radiological data (personal communication) and are anticipated to be available for microscopy data soon. The method of Hadjidemetriou et al. [1] assumes a round cell geometry, which could be directly applicable in neuroimmunology for lymphocyte lineage migration tracking [13] or for monitoring neural progenitors that retain a rounded shape [14]. Overwhelming the shape limitation to include arborized cells, such as mature neurons and astroglia, shall augment the method’s pertinency in neuroscience [15].

In the era of astonishing technical evolutions in neuroimaging and clinical neuroscience, MRI imaging, including MR-angiography, is doubtlessly at the leading edge, both in the number of investigated patients and associated costs for the healthcare system. Big data yields are associated with incidental findings, sometimes with immense clinical consequences, such as intracranial aneurysms. Tremendous material improvements and steadily growing technical experience in minimally invasive endovascular aneurysm therapy introduced a critical dilemma: “to-coil, or not-to-coil” an unruptured endocranial aneurysm. Patient consultation requires evidence and personalized yet unbiased approaches which are, to date, available as Likert scale systems, such as the PHASES score in the renounced research of Bijlenga et al. [16]. In this issue, a brand-new approach by a research group from the University of Leipzig uses artificial intelligence for aneurysm rupture risk assessment. A large (N = 446), balanced database from the authors’ medical center is engaged in a retrospective study using a gradient-boosting machine from the scikit-learn library and the R-package for statistical computing [4]. Challenging results came from this research, calling for optimistic yet cautious interpretations. Using Caucasian candidates, the authors reveal the body mass index (BMI) as the most critical rupture predictor. Aneurysm size and other comorbidities such as diabetes and hypertension, classically considered triggers for aneurysm rupture, were less critical. The machine learning approach [4] improved the aneurysm rupture prediction performance [16]. However, the receiver operating curve of classical scoring (“PHASES score”) in the databases of Bijlenga et al., and Walther et al., is not similar, which might harbor a confounding factor. The possibility of confounders does not underscore the importance of the cited clinical studies. On the contrary, it highlights the urgent necessity for public multicenter data availability in open-access databases. Over the next decade, a realistic outlook of clinical neuroscience should incorporate machine-learning model testing on multicenter, publicly available big-data collections.

3. Conclusions

In this Special Issue, we opted to motivate researchers from the wet lab, computer science, and clinical environments to join forces in a collection of highly specialized, novel achievements with a common denominator: the interdisciplinary engagement of informatics, machine learning, and deep learning. From the editor’s point of view, the core message is that interdisciplinary science shall steer and scaffold the leading experimental and clinical railroads for the next decade. Featured manuscripts highlight the urgent need for intercollected, large and curated public databases of laboratory and clinical data as a minimum prerequisite for the qualitative, robust, and comparable training of automated data analysis pipelines.