Active Learning-Enabled Brain-On-A-Chip Design for Neural Drug Development
Abstract
Brain-on-a-chip technology is advantageous to the development of personalized neural drugs, capable of finely controlling cellular microenvironments and introducing external stimuli/drugs on-demand to facilitate the neurite outgrowth. Because of multiple controllable parameters in brain-on-chips, there is an urgent need to reduce capital and time investment by creating a pre-clinical prediction model to automate the development of effective therapies. Herein, we aim to implement a hybrid approach through high-throughput brain-on-chips, wetlab experiment, and emerging machine learning (ML) tools for the construction of prediction model with high accuracy. A three-stage ML framework will be realized, including boundary definition, active learning, and data augmentation. First, the US team will collaborate with the Taiwan teams to identify influential therapeutic parameters (including growth factors, external stimuli, biointerfaces) in brain-on-chips. Through a series of feasibility tests, the upper and lower limits of each therapeutic parameter will be defined, and a support-vector machine screening layer can be thus trained. Second, the teams will continue to execute multiple active learning loops to construct a prediction model stagewise. We propose to execute 20 loops of active learning with a total of more than200 different therapies tested. Third, an in silico approach will be conducted by the US team to synthesize virtual data points by adopting various data augmentation methods followed by model optimization, which can maximize the model’s prediction accuracy. The ML-enabled prediction model can execute two tasks, including (1) achieve automatic design of customized brain-on-chips with high neurite outgrowth rates and (2) uncover complex correlations between therapeutic parameters and neurite outgrowth through data analyses. Finally, the model-suggested therapies will be validated through in vitro experiment and in vivo animal models by Taiwan labs. We believe the unique collaboration between Taiwan and US teams will produce interdisciplinary research works that have not been presented before and can provide high impacts.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Nov 17, 2022
- Source ID
- FA23862114065
Entities
People
- Po-Yen Chen
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Maryland