Machine learning and information theory help to identify the neural basis of gut feeling
A team of researchers led by Wulf Haubensak (IMP), in collaboration with VRVis, identified the basis on which decisions are made "from the gut", by analyzing data on neuronal activity.
For many people, the brain is the most complex and fascinating of all organs. The first documented brain operations took place as early as 5000 years BC. The human curiosity to understand the processes of the neural system and the brain goes back at least to the Bronze Age. Nowadays, modern neurosciences can rely on a wide range of tools to gain new insights - from imaging techniques to behavioural experiments and genetic tools. All these methods and procedures have one thing in common: they produce large, complex and heterogeneous data. In evaluating this "big data", the neurosciences are increasingly turning to methods from computer science and higher mathematics. This is precisely where VRVis' work in the field of data-driven neuroscience applies: in the implementation and development of new methods for extracting the information that rests in our neurons and brain cells, providing answers to the way the body functions.
The neural basis of gut feeling
In nature, survival often depends on making the right decision, trying something out or avoiding it. The proper behaviour here sometimes makes the difference between food security and mortal danger. These so-called affective decisions must be rooted in mechanisms that allow the brain to assign a value to its environment. Researchers have now succeeded in identifying precisely these mechanisms by combining methods from different disciplines. First, thirsty mice were conditioned to Pavlovian sounds: While they were initially unsure what the sounds meant, they learned that one of them predicted a rewarding drop of water and another a slight footstock. Successful association then evoked either approach to the water or aversive freezing responses to the sounds, indicating that the initial uncertainty on sound value was resolved with learning. The brain activities that took place during this learning process were recorded in detail. The data was then analysed using machine learning and information theory methods to map the neural network between the amygdala, the basal forebrain and the insular cortex, an area of the brain that perceives interoceptive signals.
The study shows that the brain initially does not know how to deal with new sounds and must first learn to link environmental stimuli with physical sensations in order to make behavioural decisions. For this, the amygdala sends signals to the basal forebrain, which in turn stimulates the insular cortex. As learning progresses, the sound-related activity patterns in the insular cortex increasingly resemble already known situations that reflect punishment or reward: Sounds and noises from the environment thus acquire an "intrinsic" meaning, they begin to "feel" good or bad. This value information from the insular cortex is then projected back into the amygdala, where it guides the behavioural response.
However, when the researchers blocked this information, the animals were more likely to make the wrong behavioural decision. This shows that the interaction between the insular cortex and the amygdala assigns body signals to external stimuli and uses this process to influence future behavioural decisions. On the one hand, these results explain the mechanism by which emotional decisions are made "from the gut". On the other hand, the research also shows that faulty communication between the amygdala and the insular cortex could limit the ability to behave "correctly" in uncertain situations. Low tolerance for unexpected events plays a significant role in psychiatric phenomena such as anxiety disorders or autism, suggesting a path to further research based on the study.
The communication link between the amygdala and the insular cortex
The IMP - Research Institute for Molecular Pathology - in Vienna is one of the world's leading scientific institutions in basic molecular biology research. A team led by IMP group leader Wulf Haubensak has now investigated the neuronal basis for intuitive decisions, for the "gut feeling", with mice. Together with the Biomedical Image Informatics Group of VRVis, the scientists explored the information exchange between the brain regions amygdala and insular cortex.
VRVis was responsible for the Computational Neuroscience/Data Science work in this project and supported the IMP in the computational analysis of the large amounts of data. Based on neuronal activity in the brains of mice measured with special examination methods, a network consisting of insular cortex and amygdala was discovered, which could integrate physical signals into emotional behaviour. Electrophysiological recordings and imaging techniques were used to record the activity of hundreds of neurons in this network. In order to better map the information content of this neuronal activity during the learning processes taking place in experiments, complex mathematical approaches prove suitable. Therefore VRVis applied machine learning methods and built neuronal decoders. The researchers trained artificial neural networks to predict the presence of stimuli or behaviours from the measured activity of the neurons. The accuracy of these decoders allowed statements about the extent a stimulus might be represented in a brain region.
At the same time, it was of great importance to also investigate the information flow between the different brain regions. For this purpose, VRVis used information theory methods, specifically transfer tropism. This allows measuring the flow of information in networks from complex amounts of data. By using transfer entropy, the VRVis researchers were able to calculate the information transfer from the activity of a large number of neurons. As a result, they were able to see that there is an extensive transfer of information between the cortex of the islets and the amygdala - which was the basis for correct intuitive behavioural decisions.
The use of machine learning and information theory methods (transfer entropy) was essential to extract relevant information from the large amounts of neural activity data. On this basis, the communicative bridge between the amygdala and the insular cortex, which is responsible for affective or emotional learning, was identified. This made it possible to decipher how the affective value, i.e. the emotional meaning, comes about in the neuronal activity pattern of this network.
The research work was published in the renowned scientific journal eLife.