MegaTox for analyzing and visualizing data across different screening systems

项目摘要 计算毒理学的目标是对特定的终端使用基于先前数据的规则、模型和算法， 从而能够预测一个新分子是否会有类似的负债。我们最近的努力已经 使用PubChem和ChEMBL等资源为不同的毒性相关和药物建立预测模型 发现终端。我们名为MegaTox的第一阶段SBIR提案将提供毒性机器学习模型 针对40-50个体外和体内毒性数据集开发了不同的算法。我们建议使用这一点 生成用于预测潜在化合物抗转化生长因子-a的机器学习模型的技术 对抗氯引起的肺部炎症的靶点以及识别激动剂的腺苷A1受体 作为潜在的抗惊厥药物。此外，我们还可以编辑能够重新激活乙酰胆碱酯酶的分子。 这将使发现解决神经毒剂和杀虫剂问题的医学对策成为可能 下毒了。我们将访问多种机器学习方法，并验证这些贝叶斯或其他机器 学习模型(包括线性Logistic回归、AdaBoost决策树、随机森林、支持向量 不同深度的机器和深度神经网络(DNN))，采用我们自己的内部技术 目标。我们的目标是ROC值大于0.75，MCC和F1得分是可接受的(&gt；0.3)。这些 模型将用于虚拟筛选FDA批准的药物、临床候选药物、商业可用药物 或其他分子。我们将选择多达50个分子进行体外测试，并与对照组一起进行 每个目标。这些联合努力首先应该提供商业上可行的治疗方法， 将被用来从实验上验证我们的计算模型，这些模型可以与医疗部门共享 对策科学界。总之，我们建议为目标建立和验证模型 基于公共数据库，选择要测试的化合物，创建专有数据，并以此为起点 如有需要，可进一步优化化合物。我们的目标是为每个人确定至少一种有希望的化合物 然后我们追查和保护我们的知识产权的目标。我们将寻求额外的拨款，以采取这些医疗 通过额外的体外和体内临床前研究采取对策。最终，我们将授权我们的产品 在临床试验之前交给较大的公司进行开发。

Project Summary Computational toxicology aims to use rules, models and algorithms based on prior data for specific endpoints, to enable the prediction of whether a new molecule will possess similar liabilities or not. Our recent efforts have used sources like PubChem and ChEMBL to build predictive models for different toxicity-related and drug discovery endpoints. Our Phase I SBIR proposal called MegaTox will provide toxicity machine learning models developed with different algorithms for 40-50 in vitro and in vivo toxicity datasets. We propose using this technology to generate machine learning models for predicting potential compounds against either TGF- a target for countering chlorine induced lung inflammation as well as the adenosine A1 receptor to identify agonists as potential anticonvulsants. In addition, we can also compile molecules that can reactivate acetylcholinesterase which would enable the potential to discover medical countermeasures to address nerve agent and pesticide poisoning. We will access multiple machine learning approaches and validate these Bayesian or other machine learning models (including Linear Logistic Regression, AdaBoost Decision Tree, Random Forest, Support Vector Machine and deep neural networks (DNN) of varying depth) with our own in-house technology for these selected targets. We will aim for ROC values greater than 0.75 and MCC and F1 scores that are acceptable (>0.3). These models will be used to virtually screen FDA approved drugs, clinical candidates, commercially available drugs or other molecules. We will select up to 50 molecules to be tested using in vitro assays alongside controls for each target. These combined efforts should in the first instance provide commercially viable treatments which will be used to experimentally validate our computational models that can be shared with the medical countermeasures scientific community. In summary, we are proposing to build and validate models for targets based on public databases, select compounds for testing, create proprietary data and use this as a starting point for further optimization of compounds if needed. Our goal is to identify at least one promising compound for each target that we then pursue and protect our IP. We will pursue additional grant funding to take these medical countermeasures through additional in vitro and in vivo preclinical studies. Ultimately, we will license our products to larger companies for development prior to clinical trials.