Efficient mapping of neuromodulation therapy parameters to complex biomarker spaces

引言：神经调节领域在转换疗法方面已经取得了许多早期的成功 进入临床应用，但在减少新适应症的脱靶效应方面仍然存在挑战， 以及靶向治疗，以确保临床和商业可行性。挑战已经出现 由于临床前和临床试验之间的差异， 导致需要不同刺激参数的临床受试者。这些挑战 阻碍了神经调节扩展到新适应症[Herring 2019]。在此，我们旨在 展示了如何利用神经信号中的信息来量化和避免脱靶 效果[Ardell 2017]为未来治疗提供更好的特异性。 项目概述：我们建议利用现有的数据库数据建立一个有效的方法， 使用高斯贝叶斯优化搜索刺激参数空间 过程作为模拟刺激响应函数的数据有效方式。此外，委员会认为， 我们将建立一个机器学习模型，用于将神经信号映射到压缩的 神经生物标志物（Neural Biomarkers），以将身体的神经信号解释为 器官功能的有效措施。最后，我们验证了这两种方法在急性和慢性 通过输入广泛搜索刺激参数的大型动物模型 空间，并针对源自神经信号的神经生物标志物进一步优化。

Introduction: The neuromodulation field has seen many early successes in converting therapies into clinical application, but challenges remain in reducing off-target effects for new indications as well as targeting therapies to ensure clinical and commercial viability. Challenges have arisen validating preclinical successes in clinical settings due to the differences between preclinical and clinical subjects leading to different stimulation parameters required. These challenges have impeded the expansion of neuromodulation to new indications [Herring 2019]. Herein we aim to show how the information in the neural signals can be leveraged to quantify and avoid off-target effects [Ardell 2017] providing better specificity for future therapies. Outline of the project: We propose utilising existing SPARC data to build an efficient method of searching the stimulation parameter space using Bayesian Optimisation with Gaussian Processes as a data-efficient way of modelling the stimulation response function. Furthermore, we will build a machine learning model for mapping neural signals into a compressed representation (termed Neural Biomarkers) in order to interpret the body’s neural signals as effective measures of organ function. Finally, we validate both approaches in acute and chronic large animal models where stimulation parameters are searched extensively through the input space and further optimised against neural biomarkers derived from the neural signals.