Creating AI/ML-ready data for single cell proteomics

项目总结：补充1R01GM147653-01 Brief Parent R01摘要 生命是发生在单个细胞内部和之间的动态相互作用的结果。单细胞 分析表征了样本的多样性以及每个细胞的状态和反应能力 对环境的影响。单细胞蛋白质组学(SCP)正在迅速崛起，可以量化&gt；1000 每个细胞的蛋白质，足以对细胞类型进行分类和揭示特征的覆盖水平 细胞功能。仪器和样品制备方面正在取得重大进展 更广泛地使用SCP。然而，数据采集方面的进步并没有与 先进的计算工具。 父奖创建专门针对SCP数据的独特性质进行优化的算法， 并将从根本上提高单细胞蛋白质组的准确性和覆盖率。该计划的目标是 家长奖解决了频谱识别(目标1)和蛋白质方面的算法挑战 量化(目标2)。该项目还创建了一些单细胞蛋白质组学数据 标杆目的(目标3)。 《补编》的目标 补编提案是家长奖的MPI和专家之间的合作 在机器学习中。补充的目标是改善单细胞的AI/ML准备情况 蛋白质组学数据通过两个主要任务。首先，我们将定义和实现以下文件格式 比蛋白质组学数据和结果的格式更容易接受机器学习，因为 当前格式臃肿且不足，无法扩展到必要的数据级别 机器学习所需的。第二，我们将改进软件工具，以捕获 元数据，这是描述实验设计所必需的。随着成功的 完成这两项任务，我们将演示来自单细胞蛋白质组学的数据 实验可立即用于各种机器学习任务，可在 ProteomicsML.org。

Project Summary: Supplement to 1R01GM147653-01 Brief parent R01 summary Life is the result of dynamic interactions that occur within and among individual cells. Single cell analyses characterize a sample’s diversity and each individual cell’s state and ability to respond to the environment. Single cell proteomics (SCP) is rapidly emerging and can quantify > 1000 proteins per cell, a level of coverage sufficient to categorize cell types and reveal characteristic cellular functions. Significant advances in instrumentation and sample preparation are making SCP more broadly accessible. Yet advances in data acquisition have not been paired with advances to computational tools. The parent award creates algorithms specifically optimized for the unique nature of SCP data, and will radically improve accuracy and coverage of the single cell proteome. The aims of the parent award address algorithmic challenges in spectrum identification (Aim 1) and protein quantification (Aim 2). The project also creates some single cell proteomics data for benchmarking purposes (Aim 3). Goals of the Supplement The Supplement proposal is a collaboration between the MPIs of the parent award and experts in machine learning. The goal of the supplement is to improve the AI/ML readiness of single cell proteomics data through two primary tasks. First, we will define and implement file formats that are more amenable to machine learning than the formats for proteomics data and results, as current formats are bloated and insufficient and cannot scale to the necessary level of data required for machine learning. Second, we will improve software tools for the capture of meta-data, which is essential for describing the experimental design. With the successful completion of these two tasks, we will demonstrate that data from single cell proteomics experiments are immediately usable for a variety of machine learning tasks available at proteomicsML.org.