Drug discovery and development for central nervous system disorders face unique challenges and in general have higher failure rates compared with other disease areas. Central nervous system disorder is a broad category of conditions in which our brain does not function as it should, resulting in the limitation of people’s health and brain function. Although tremendous efforts have been paid to study CNS diseases, the pathophysiology of some diseases, such as Alzheimer’s Disease (AD), and Parkinson’s Disease (PD), remain poorly understood by people. The mammalian central nervous system is highly complex and responsible for coordinating and influencing most functions of the body. Many things could go wrong in many ways. Therefore, novel approaches are highly demanded, not only to better understand brain function and physiology, but also to accelerate the development for CNS drugs. Whole organism brain activity mapping has been proved to be an invaluable step toward understanding fundamental and pathological brain processes. However, brain activity mapping must address some leading issues, such as the experimental throughput, also the temporal and spatial resolution of acquired data. The advancement of microscopy technology and protein engineering is of significant help. On the other hand, what we cannot ignore is how to analyze multimodal brain activity map data accurately, comprehensively, and efficiently. Artificial intelligence is experiencing a skyrocketing development and has been employed into various areas.

In this dissertation, we present the application of advanced neural engineering techniques, including microfluidic system, multielectrode electrophysiological recording, and advanced imaging technology, to acquire massive multimodal brain activity map data and study brain function and connection. With the enhancement of artificial intelligence algorisms, massive data could be deconvoluted and this would be a promising way to interpret the mechanisms underlying various brain conditions and recognize the drug therapeutic pattern.

The dissertation begins with an introduction to current status of drug discovery for CNS disorders, which elaborates on the traditional development and major challenges of the CNS drugs. Further, it introduces emerging technologies for improving and accelerating CNS drug development, like high-speed microscopy, advanced fluorescent indicators, and artificial intelligence algorisms, which would be a future direction to explore brain physiology. In chapter 2, we refined our previous HT-BAMing platform with a larger training set and a deep learning algorism. The advancement of imaging technology and high-throughput experiment platform has led us to a big data era, and interpreting data comprehensively and efficiently grows to be a challenging task. Herein, based on previous experience in the microfluidic system design and high-throughput drug screening, we introduced a high-throughput in vivo DeepBAM drug screening platform, which enables massive brain activity map collection and drug mechanism prediction. Microfluid-chip and high-speed microscopy are utilized in DeepBAM to generate large-scale brain activity data from drug-treated zebrafish larvae. Accordingly, from a screen of clinically used drugs, we build a functional classifier based on deep learning, which employed the convolutional neural network to extract representative features. Then the high-level features from clinical compounds are used to train a random forest classifier and non-clinical compounds are applied to predict the corresponding mechanisms. As a result, several potent anti-Parkinson’s and anti-epilepsy drugs were found from the 121 non-clinical drugs with literature support. Collectively, our DeepBAM platform would be of significant value in large-scale drug screening and mechanism-unknow drug prediction, dramatically reducing the traditional complex drug development processes.

DeepBAM platform exhibits versatile performance over the prediction of therapeutic-unknown compounds and successfully predicts several potent anti-Parkinson’s drug candidates. However, all collected data are in the 2D overlapped format, which means that we could only decode the information from a zipped, downscaled plane. The brain is known as a highly complicated 3D organism with multiple brain regions responsible for various functions. In Chapter 3, to tackle this limitation, we proposed an AI-functional brain activity platform enabled by advanced light-sheet microscopy and machine learning algorism. Our platform, for the first time, facilities large-scale 3D functional brain activity mapping with single-cell resolution. We have collected information-rich spatial brain activity mapping data in awake zebrafish larvae after treatment of 86 compounds of interest. we succeeded in addressing the problem of 3D image alignment and registration, which is one tough and complex task in 3D image analysis. Accordingly, we conduct bioinformatic analysis on the functional brain activity data, to study the whole-brain-wide physiology and recognize specific drug therapeutic patterns. In this regard, our AI-functional brain activity mapping platform bears great merits for both 3D brain activity mapping in larval zebrafish exposure to external stimulation, and AI-enhanced drug pattern recognition.

This dissertation ends with Chapter 4, which summarizes our development in the current study as well as the challenges, and also presents a perspective on possible directions for future investigation.

Ultimately, with the enhancement of neural engineering and artificial intelligence techniques, we successfully documented three versatile platforms that realize a multimodal brain activity map acquisition, evaluation, and pattern recognition. We deeply believe that these strategies will be of significant help in exploring brain physiology and recognizing specific system neuropharmacology for discovering novel CNS drugs.