High Precision Classification of Neuronal Polarity in Drosophila Brain by using Deep Neural Network

Kuan-Ting Chou^1,2*, Caroline Su^1,2, Chung-Chuan Lo^2,3, Daw-Wei Wang^1,2,4

¹Physics Department, National Tsing Hua University, Hsinchu, Taiwan
²Brain Research Center, National Tsing Hua University, Hsinchu, Taiwan
³Institute of Systems Neuroscience, National Tsing Hua University, Hsinchu, Taiwan
⁴Physics Division, National Center for Theoretical Sciences, Hsinchu, Taiwan

* Presenter:Kuan-Ting Chou, email:stu95.40801@gmail.com

Identifying the direction of signal flows in neural circuits is the most important step for understanding the intricate dynamics of a brain. In principle, this information can be identified by experimentally measuring the dendrite-axon polarity of each neuron. However, such direct measurement cannot be applicable to existing neuronal databases in which polarity information is not fully identified. Here we construct a machine learning model with Deep Neural Network (DNN) to identify the neural polarity (i.e. dendrite or axon) in the Drosophila Brain. Using skeleton structure of 67 labelled neurons from the Flycircuit, our model has provided a very high precision and recall rates (> 95%). After further training with more available data, our DNN model should be capable to quickly and accurately identify the polarity of all the known neurons in the Drosophila brain.

Keywords: deep learning network, neuronal polarity