LMB 2023/07 Machine Learning GPU Cluster

LMB 2023/07 机器学习 GPU 集群

• 批准号: MR/Y001966/1
• 负责人: Jake Grimmett
• 金额: $ 91.66万
• 依托单位: MRC Laboratory of Molecular Biology
• 依托单位国家: 英国
• 项目类别: Research Grant
• 财政年份: 2023
• 资助国家: 英国
• 起止时间: 2023 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=MR%2FY001966%2F1
• 关键词: LMB; 2023; 07; Machine; Learning

Machine Learning, also known as Artificial Intelligence, is a powerful new computational method, that has recently shown itself to be extremely useful in Biological research. A prominent example is AlphaFold-2; a tool for predicting protein structures based on primary sequence. This tool so much improved on previous computational methods for predicting protein structures, that it can be genuinely termed groundbreaking. Other structural biological problems, such as the prediction of protein-protein interactions have similarly benefited from the application of machine learning tools, as too have related problems, such as automated atomic modelling in cryo-EM maps. Machine Learning is not restricted to solving structural problems, it can extract phylogenetically useful patterns from amino acid or nucleotide sequences, and even predict protein function from similar data. Aside from sequence based data, machine learning can effectively analyse images. For example, machine learning can trace the path of neural axons through a 3D stacks of images. This will make the task of generating a complete wiring map of a vertebrate brain finally possible - it would otherwise take 1000's of human work years to complete. Similarly, machine learning can be used to analyse timecourse photos of embryo development, and track both individual cells, and cell lineages. The LMB is both a user, and developer of Machine Learning methods. As Machine Learning is an emergent, rapidly developing technology, we require a small, state of the art machine learning cluster. We propose to invest in two of the latest generation of Machine learning system, each with eight Nvidia H100 80GB SXM GPUs, interlinked with 200Gb ethernet.

机器学习，也被称为人工智能，是一种强大的新计算方法，最近已被证明在生物研究中非常有用。一个突出的例子是AlphaFold-2;一种基于一级序列预测蛋白质结构的工具。这个工具大大改进了以前预测蛋白质结构的计算方法，可以说是真正的突破性。其他结构生物学问题，如蛋白质-蛋白质相互作用的预测，也同样受益于机器学习工具的应用，相关问题也是如此，如冷冻EM图中的自动原子建模。机器学习并不局限于解决结构问题，它可以从氨基酸或核苷酸序列中提取遗传学上有用的模式，甚至可以从类似的数据中预测蛋白质功能。除了基于序列的数据之外，机器学习还可以有效地分析图像。例如，机器学习可以通过3D图像堆栈跟踪神经轴突的路径。这将使生成脊椎动物大脑的完整布线图的任务最终成为可能-否则需要1000年的人类工作才能完成。同样，机器学习可以用于分析胚胎发育的时间过程照片，并跟踪单个细胞和细胞谱系。LMB既是机器学习方法的用户，也是开发者。由于机器学习是一种新兴的、快速发展的技术，我们需要一个小型的、最先进的机器学习集群。我们建议投资两个最新一代的机器学习系统，每个系统都有八个Nvidia H100 80 GB SXM GPU，与200 Gb以太网互连。