Deep Learning Behavioral Phenotyping System in the Diagnosis of Parkinson's Disease with Drosophila melanogaster

Drosophila Melanogaster is widely used as animal models for Parkinsons disease (PD) research. Because of the complexity of MoCap and quantitative assessment among Drosophila Melanogaster, however, there is a technical issue that identify PD symptoms within drosophila based on objective spontaneous behavioral characteristics. Here, we developed a deep learning framework generated from kinematic features of body posture and motion between wildtype and SNCAE46K mutant drosophila genetically modeled {square}-Syn, supporting clustering and classification of PD individuals. We record locomotor activity in a 3D-printed trap, and utilize the pre-analysis pose estimation software DeepLabCut (DLC) to calculate and generate numerical data representing the motion speed, tremor frequency, and limb motion of Drosophila Melanogaster. By plugging these data as the input, the diagnosis result (1/0) representing PD or WT as the output. Our result provides a toolbox which would be valuable in the investigation of PD progressing and pharmacotherapeutic drug development.