My research in computational neuropsychiatric genomics on the zebrafish has lead me to investigate what sort of methods and inquiries I could put forward using statistics and algebra. Though zebrafish are an inherently helpful model organism for psychiatric disease I want to extend the nature of zebrafish research such that we can achieve the full potential of psychiatric disorders no matter what species we are studying. The results of artificial intelligence in particular hold promising techniques that extend into biology and neuroscience. As scientists peek into the architectures and algorithms like hierarchical filtering and supervised learning, they can create more detailed and elaborate explanations of biological phenomena. OpenSource platforms in particular need to establish a framework or fundamental principles by which scientists can draw conclusions on the nature of psychiatric disease itself through accounting for the limits of experimental observation. In this post, we’ll discover some of the latest findings in computational neuroscience as they relate to the questions we’d like to answer.
As the field of neuroscience reaps the benefits brought upon by big data and advances in next generation sequencing, scholars have raised issues and questions brought upon by the questions and challenges they wish to address. Modeling data itself is a very important step for scientists to understand the nature of disease and neuroscientific phenomena. To benefit a field like psychiatry, researchers must find models formulated through prose that rely on several available empirical findings. But psychiatry and neuroscience can borrow principles and techniques from physics and mathematics, particularly those ideals which seek precise and explicit methods of creating predictions. The precision used in characterizing gravitational waves and in discovering the existence of the Higgs Boson are exceptional examples which show promising results for extending these processes into neruoscience. Still, the available models in computational neuroscience and neuropsychiatric genomics remain difficult to falsify and more descriptive than predictive. We see a coming together of disciplines, but not exactly the promising results that we had anticipated. For this reason, scientists must develop precise models drawing from modern neuroimaging techniques and mathematics.
On the bright side, neuroimaging has shown how complex, interconnected brain changes characterize psychiatric illnesses. This has informed researchers to select treatment targets of new therapies and form predictions on genetic risks for patients (Mayberg, et al. 2005). Many models remain lacking the appropriate mathematics to justify their use, however. Further statistical tests such as cross-validation would be necessary to justify generalizing a linear model from a set of training and test data. Computational researchers have put forward efforts, though, as shown in the issue of Biological Psychiatry: Cognitive Neuroscience and Neuroimaging. The issue showcases techniques from mathematics, physics, and engineering used in creating more rigorous models of psychiatric disorders. The articles in the issue share the premise that brain connectivity is the cause of psychiatric disorders, and, from this, scientists are able to borrow from graph theory, network theory, geometry, and even topology in describing networks and their functions. The tools are certainly available to researchers, but putting them into context and using them the appropriate way remain to some degree elusive. Network science (such as the work of Morgan et al., 2018) in recent months has shown effective results in studying autism, ADHD, and psychosis. The methods of borrowing techniques from one discipline to another require constant validation with the empirical evidence surrounding psychiatric illness to prove their validity.
Other advancements (such as that of Scholtens et al., 2018) search for unifying principles that can be used to explain any (or at least, many) data. These approaches account for the differences among varying sets of data such that the equations and algorithms that use network theory can be easily generalizable to any issue. Scientists should focus on these fundamentally-driven, elegant solutions to uncover the multiscale complexity of brain function. This method of drawing simplicity from complexity and moving between different levels of function and organization are what would allow neuroscience and neurobiology to actualize the full potential of their disciplines much the same way mathematics and physicists function as well.
Moving from descriptive to predictive models, as well, is a bottleneck for many models, and some researchers (like Janssen et al., 2018) have laid the foundation for using machine learning techniques to perform these predictions. The complexity and sheer amount of neuroimaging data can be simplified (in accordance with appropriate computational and physical limitations) and used for prediction illness outcome. They authors do warn, though, that overestimation of prediction performance is possible as test data may be overfit to training data. These machine learning methods still need to account for the multidimensional nature of large data sets and create realistic measurements of performance effectiveness.
Researchers need to engage in further work in determining the soundness and benefits of these models, especially from basic science principles to the clinical setting, to yield the true potential of computational neuroscience. For the complexity of psychiatry and biology to meet the same certainty and precision of mathematics and physics requires constant evaluation of scientific techniques. While the work done in recent years is a good start, there’s much, much more to be discovered.
Janssen, R.J., Mourão-Miranda, J., and Schnack, H.G. Making individual prognoses in psychiatry using neuroimaging and machine learning. Biol Psychiatry Cogn Neurosci Neuroimaging. 2018; 3: 798–808
Mayberg, H.S., Lozano, A.M., Voon, V., McNeely, H.E., Seminowicz, D., Hamani, C. et al. Deep brain stimulation for treatment-resistant depression. Neuron. 2005; 45: 651–660
Morgan, S.E., White, S.R., Bullmore, E.T., and Vértes, P.E. A network neuroscience approach to typical and atypical brain development. Biol Psychiatry Cogn Neurosci Neuroimaging. 2018; 3: 754–766
Scholtens, L.H. and van den Heuvel, M.P. Multimodal connectomics in psychiatry: Bridging scales from micro to macro. Biol Psychiatry Cogn Neurosci Neuroimaging. 2018; 3: 767–776
Leave a Reply