Neural algorithms underlying diversity in visual feature integration

Animals are capable of selecting diverse behaviors in response to highly similar sensory inputs. Behavioral output diversity then depends on the neural algorithms employed by circuits to transform the sensory information into behavioral outputs. The mechanisms that underlie these transformation processes, however, remain primarily unknown. In a collaboration with Dr. Catherine von Reyn's group we use looming feature integration in Drosophila melanogaster as a well-defined model of sensory-motor integration to uncover the neural algorithms that transform visual features into multiple distinct, behaviorally relevant outputs.

We integrate modern machine learning and data science techniques with biophysically realistic computational models of Drosophila neurons and develop formal descriptive mathematical models to directly derive and verify sensorimotor integration algorithms computed in the neuronal pathways that transform visual information into motor programs.

Our aim is to reveal novel algorithms and encoded sensory features that are currently missing in the field and clearly identify how different algorithms play together or contrast each other to achieve output diversity across different descending pathways. These algorithms can then be used as an anchor across the animal kingdom to contrast and compare underlying mechanisms and dynamics. Together, these data will provide great insight into how behavioral diversity is generated in sensory evoked behaviors across species, and in general across sensorimotor transformations. 

Control strategies of descending command systems for the generation of context-specific locomotor behaviors

Locomotion represents the final output of the central nervous system and needs to be continuously adapted to the task at hand. Centers in the brainstem transform inputs from higher motor regions into meaningful commands to control speed and gait, stop or freeze locomotion, direct turning maneuvers, and orchestrate the transitions between them. Many of these behaviors might be organized and controlled by distinct neuronal subsets and pathways. In addition to their role in initiating, driving and terminating different behaviors, these brainstem areas have also been implicated in recovery after spinal cord injury and represent promising targets for deep brain stimulation in patients with Parkinson’s disease. However, to understand pathologies of motor behavior and to open avenues for partial or complete restoration of function, a thorough understanding of the underlying neuronal circuitry is required.

Despite recent progress in unraveling the neuronal populations and pathways involved in supraspinal control of locomotion, the logic of the processing in these centers and the transformation of their output into locomotor activity is still unclear. We therefore develop detailed data-based computational models to systematically structure, integrate and probe current knowledge on the organization and function of subpopulations of the mesencephalic locomotor region and the reticulospinal system. Our models are valuable complements and facilitators for further experimental investigations, identifying key strategies in locomotor control and providing a theoretical framework for the future investigation of motor control strategies and restoration of function after disease or injury.