SMARTCORE Technology: Using AI and Patient Tissue to Identify Potential Cancer Therapies for Ultra-rare Cancers

项目摘要 我们的提案提出了一种名为SmartCORE的新方法，以应对寻找实际解决方案的挑战 治疗极度罕见的癌症。我们的目标是使用人工智能驱动的药物筛选平台来测试人类原发肿瘤 组织。我们的战略是识别和重新定位那些对感兴趣的癌症具有活性的药物。我们 通过以下方式避免开发用于药物筛选的有机物或患者来源的异种移植模型的潜伏期 利用完整的肿瘤作为器官培养物。使用机器学习算法，我们可以预测肿瘤 对一组约4,000种药物的敏感性，包括约1,800种FDA批准的药物，通过输入对一组 通过计算选择了254种化合物。我们的长期目标是创造一个强有力的药物和靶点发现 用于个体癌症的方法，无论是否罕见，都直接与临床应用相联系，从而解决 在个性化和精确化肿瘤学中，一个关键的“床到床”缺口。为了实现我们的目标，我们打算扩大 我们平台利用临床针吸活检的能力。这将极大地提高我们的 SmartCORE技术可以分析可以运往我们处理设施的任何肿瘤的活检样本 在24小时内。此建议书将以极其罕见的方式展示我们的SmartCORE平台的概念验证 条件，肝纤维板层癌，有两个特定的目标。目标1是发现治疗方法 纤维板层癌的人工智能化学筛查。在这里，我们将建立最优的技术参数 我们基于人工智能的筛查平台使用从多个来源获得的独立FLC队列的性能， 使用来自三个独立的人类FLC的已建立的患者来源的异种移植模型来验证排名靠前的异种移植模型， 并推断候选药物靶向的信号网络和蛋白质，突出分子途径 对FLC的生存至关重要。目标2是开发一种基于人工智能的使用针头的化学筛选方法 活组织检查。为了扩大我们技术的影响，我们将修改我们的基于人工智能的筛选，以适应18- 在临床实践中常规进行的计量芯针活检。我们将为FLC策划一套40种药物-- 具体测试作为药物预测的基础，使用我们的深度神经网络算法，测试这一方法的准确性 通过比较我们FLC人群中的针吸活检和较大组织切片的方法并优化分析前 产生可重现结果的条件。如果成功，我们的技术将利用针芯活检， 对潜在的分子错乱不可知，使用深层神经筛选大量化合物集合 网络算法，从开始到结束需要一周的短时间；所有这些都是属性 克服个性化肿瘤学的致命弱点的理想测试。

Project Summary Our proposal presents a new approach, called SmartCore, to tackle the challenge of finding practical solutions for ultra-rare cancers. We aim to use an AI-driven drug screening platform designed to test primary human tumor tissue. Our strategy is to identify and repurpose drugs that exhibit activity against the cancer of interest. We circumvent the latency of developing organoids or patient-derived xenograft models for drug screening by utilizing intact tumors as organotypic cultures. Using a machine-learning algorithm, we can predict tumor sensitivity to a panel of ~4,000 drugs, including ~1800 FDA-approved drugs, by inputting responses to a set of computationally selected 254 compounds. Our long-term goal is to create a robust drug and target discovery method for individual cancers, irrespective of rarity, that directly links to clinical application, thereby addressing a critical ‘bench-to-bedside’ gap in personalized and precision oncology. To realize our goal, we intend to expand the capability of our platform to make use of clinical needle biopsies. This will vastly increase the utility of our SmartCore technology to analyze biopsy samples from any tumor that can be shipped to our processing facility within 24 hrs. This proposal will demonstrate the proof-of-concept of our SmartCore platform in an ultra-rare condition, fibrolamellar cancer of the liver, with two specific aims. Aim 1 is to discover therapeutics for fibrolamellar cancer using AI-based chemical screening. Here, we will establish technical parameters for optimal performance of our AI-based screening platform using independent FLC cohorts obtained from multiple sources, validate the top ‘hits’ using established patient-derived xenograft models from three independent human FLCs, and deduce signaling networks and proteins targeted by the candidate drugs, highlighting molecular pathways important for the survival of FLC. Aim 2 is to develop an AI-based chemical screening approach using needle biopsies. To broaden the impact of our technology, we will modify our AI-based screening to accommodate 18- gauge core needle biopsies routinely performed in clinical practice. We will curate a set of <40 drugs for FLC- specific testing as the basis for drug prediction using our deep neural network algorithm, test the accuracy of this approach by comparing needle biopsies with larger tissue slices in our FLC population and optimize pre-analytic conditions to yield reproducible results. If successful, our technology will make use of needle core biopsies, be agnostic to the underlying molecular derangement, screen a large collection of compounds using deep neural network algorithms, and require a short turnaround time of one week from start to finish; all of which are attributes of an ideal test that overcomes the Achilles heel of personalized oncology.