AI-informed Signaling Factor Design for in vitro Rejuvenating Mesenchymal Stromal Cells

摘要 虽然间充质基质细胞(MSCs)在治疗许多具有挑战性的疾病方面有着巨大的前景 疾病，阻碍有临床意义的骨髓间充质干细胞治疗的主要障碍是无法产生强大的骨髓间充质干细胞 始终如一。具体地说，体外培养的MSCs通常会迅速进入衰老状态，在衰老过程中它们会失去效力。 与体内的自然衰老不同，这种体外衰老已被证明在很大程度上是由错误的调控所驱动的 培养中的代谢信号。为了应对这一重大挑战，许多信号通路(例如，成纤维细胞生长因子、ATM、SRT、 MTOR、EGF、DDR2)已被确定用于调节衰老相关过程。建立在这些基础上 该R35 Mira提案旨在开发一种创新的工程方法来延迟MSC 通过集体调节这些信号通路来完成衰老过程。具体地说，我们假设 经过充分训练的人工智能模型可以预测有效减缓甚至减缓的信号因子组合 逆转衰老相关的转录漂移。为了实现这一目标，我的研究旨在解决三个问题 MSC工程中的知识/技术差距(图1B)：1)如何准确地对活的MSCs进行表型(例如， 特征、增殖和效力)；2)如何预测决定所需的信号因素 转录反应；以及3)如何确保这种预测的稳健性。 在挑战1中，该提案将通过开发方法来扩展我们以前开发的人工智能平台 获取覆盖广泛MSC表型的大规模AI训练数据并解释黑盒 深度学习模型。其目的是破译骨髓间充质干细胞的形态-基因表达关系。在……里面 挑战2，我们将利用深度学习来识别信号因子组合并预测性地调整基因 骨髓间充质干细胞的表达。在第三个挑战中，我们将开发提高AI模型健壮性的算法 并将我们概念验证的人工智能平台转变为实际临床应用的可靠工具。最直接的 我们提出的研究成果将带来一个高通量的MSCs表型和工程平台。 建议的实验平台也将使我们能够更好地理解MSC 机制生物学和衰老信号的相互作用。

ABSTRACT While mesenchymal stromal cells (MSCs) hold enormous promise for treating many challenging diseases, a major barrier toward clinically meaningful MSC therapies is the inability to produce potent MSCs consistently. Specifically, in vitro cultured MSCs often rapidly enter senescence in which they lose their potency. In contrast to natural in vivo senescence, such in vitro aging has been shown to be largely driven by misregulated metabolic signaling in culture. To address this grand challenge, many signaling pathways (e.g., FGF, ATM, SRT, mTOR, EGF, DDR2) have been identified for regulating senescence-related processes. Building upon these discoveries, this R35 MIRA proposal aims to develop an innovative engineering approach to delaying the MSC senescence process by collectively adjusting these signaling pathways. Specifically, we hypothesize that a sufficiently trained AI model can predict the signaling factor combination that effectively slows down or even reverts the senescence-related transcriptional drift. To achieve such a goal, my research aims to address three knowledge/technology gaps in MSC engineering (Fig. 1B): 1) how to accurately phenotype live MSCs (e.g., characteristics, proliferation, and potency); 2) how to predict signaling factors that dictate the desired transcriptional response; and 3) how to ensure the robustness of such predictions. In challenge 1, this proposal will expand our previously developed AI platform by developing approaches to acquiring large-scale AI training data that cover a wide range of MSC phenotypes and interpreting black-box deep learning models. The goal is to decipher the morphology-gene expression relationship in MSCs. In challenge 2, we will utilize deep learning to identify the signaling factor combination and predictively adjust gene expression in MSCs. In the third challenge, we will develop algorithms that improve the robustness of AI models and turn our proof-of-concept AI platforms into reliable tools for practical clinical utilizations. The immediate outcome of our proposed research will lead to a high-throughput phenotyping and engineering platform of MSCs. The proposed experimental platform will also enable us to establish better understandings in MSC mechanobiology and senescence signaling interactions.