Deep-Learning-Aided Three-Dimensional Direct Laser Writing of the Complete Connectome of Mushroom Body from an Insect Brain

Abstract

Creation of biomimetic neuronal structures that emulate the topology of biological neural networks (BNNs) has been an active area of research in engineered neural networks over the last decade, these biomimetic neuronal structures simulate the microenvironment that the neurons could adhere, proliferate and differentiate as well as the drug release through temporal and spatial control [1]. Owing to the capability of three-dimensional processing and the fabrication resolution down to nanometers, three-dimensional (3D) direct laser writing (DLW) based on multi-photon absorption [2,3], has been widely studied and utilized to produce 3D biomimetic structures, such as biomimetic human ovarian malignancy structures and single biomimetic neuron structures [4,5]. However, large scale biomimetic neuronal structures couldn't be facilely achieved using 3D DLW due to the structural complexity of BNNs. This challenge can be articulated into two aspects: i. Unachievable fabrication time. BNNs are 3D networks structures spinning in a very large scale, take human brain as an example, the total neuron circuit length is estimated to be 100 billion μm, resulting in a fabrication time of 1 billion seconds (roughly 31.7 years); ii. Structural data of neuronal circuit at synaptic resolution are rarely available from neuroscience.