Virtual nanostructure simulation (VINAS) portal

纳米材料，特别是工程纳米材料（ENM）在消费品中的应用 在过去的十年里，产品和药品的价格一直在飙升。各种体外和体内 评估ENM潜在的环境和健康影响的研究已经产生了大量的 大量的实验数据，需要紧急管理以提取信息， 分析/建模和使用人工智能方法的数据/模型共享。计算 建模方法，特别是机器学习和深度学习方法， 期望根据现有的ENM开发预测模型， 性质/活性/毒性数据。目前ENMS数据库不包含纳米结构 注释，以机器可读格式存储各种结构信息， 用于计算机建模研究。为了应对当前大数据时代的挑战，我们将 开发一个大型的、公开可用的ENMs门户网站，其中包含更多注释的纳米结构， 超过3,000个ENM适合于计算建模研究，这将导致合理的 纳米医学设计纳米技术健康影响研究（NHIR） 联盟正在为该门户网站的初始ENMs数据库提供高质量的ENMs数据， 还将支持未来的数据管理。该数据库将基于虚拟 纳米结构模拟（VINAS）技术，将注释复杂的纳米结构 转换成适合于机器学习建模目的的机器可读格式。到 为此，我们将开发各种新的计算方法来注释纳米结构， 特别是对于复杂的ENM（例如石墨烯衍生物）。之后，我们将使用新机器 学习和深度学习算法，如加性模型和可解释的人工智能引导的半 监督式深度学习技术，使用ENM数据开发预测模型 策展数据库作为概念验证。例如，虚拟纳米材料投影 基于深度学习的方法，特别是可解释的人工智能引导的半监督 生成对抗网络，将特别擅长处理注释 纳米结构。在VINAS数据库门户网站上，作为最终交付成果， 生物活性/性质/毒性数据和注释的纳米结构将作为可下载的共享 文件供公众使用。由此产生的新的深度学习预测模型将是 也分享了。该研究为未来数据驱动的纳米信息学提供了一个新的公共平台 建模研究，特别是那些基于机器学习的方法，可以大大 推进合理的纳米医学设计和现代纳米信息学的其他领域。

The use of nanomaterials, especially Engineered Nanomaterials (ENMs), in consumer products and medicine has been skyrocketing over the past decade. Various in vitro and in vivo studies evaluating the potential environmental and health effects of ENMs have generated vast quantities of experimental data, requiring urgent curation for information extraction, analysis/modeling, and data/model sharing using artificial intelligence methods. Computational modeling methods, especially machine learning and deep learning approaches, bear high expectations to develop predictive models for ENMs based on the available property/activity/toxicity data. Currently ENMs databases do not consist of nanostructure annotations to store diverse structural information in machine readable formats that are critical for computational modeling studies. To address this challenge in the current big data era, we will develop a large, publicly available ENMs portal that contains annotated nanostructures of more than 3,000 ENMs suitable for the computational modeling research, which will lead to the rational nanomedicine design. The ongoing Nanotechnology Health Implication Research (NHIR) consortium is providing high quality ENMs data for the initial ENMs database of this portal and will also support future data curations. This database will be designed based on Virtual Nanostructure Simulation (VINAS) technique, which will annotate the complex nanostructures into machine readable formats that are suitable for the machine learning modeling purpose. To this end, we will develop various new computational approaches to annotate the nanostructures, especially for complex ENMs (e.g. graphene derivatives). After that, we will use new machine learning and deep learning algorithms, such as additive model and explainable AI guided semi- supervised deep learning technique, to develop predictive models using the ENMs data of the curated database as the proof of concept. For example, a virtual nanomaterial projection approach that is based on deep learning, particularly the explainable AI guided semi-supervised generative adversarial networks, will be especially adept at handling the annotated nanostructures. In the VINAS database web portal as the final deliverables, the curated ENMs- bioactivity/property/toxicity data and annotated nanostructures will be shared as downloadable files for public community to use. And the resulting new deep learning predictive models will be shared as well. This study provides a new public platform to future data-driven nanoinformatics modeling studies, especially those machine learning based approaches, and can greatly advance the rational nanomedicine design and other areas of modern nanoinformatics.