一个整体光谱预测 (ESP) 模型用于代谢物注释
Xinmeng Li1, Yan Zhou Chen1, Apurva Kalia1
1Department of Computer Science, Tufts University, Medford, MA, 02155, United States.
Bioinformatics (Oxford, England)
|August 24, 2024
概括
一个新的机器学习模型,集谱预测 (ESP),通过结合多层感知器和图形神经网络模型来改进代谢物注释. 这提高了在生物样本中识别分子的光谱预测准确度.
科学领域:
- 代谢学 代谢学 代谢学
- 计算化学的计算化学
- 生物信息学是一种生物信息学.
背景情况:
- 代谢学研究面临着一个重大挑战,即用精确的化学标识标注测量频谱,目前只能识别出一小部分频谱.
- 现有的代谢物注释的计算方法包括将候选分子映射到光谱或查询光谱到分子候选物,最佳匹配的光谱确定目标分子.
- 有限的研究专注于结合等级学习任务,以提高在光谱注释中确定目标分子的准确性.
研究的目的:
- 开发一种新的机器学习模型,即集谱预测 (ESP),以提高代谢物注释在代谢学中的准确性.
- 改进从生物样本中测量的光谱赋予化学标识的过程.
- 利用和结合现有的基于神经网络的注释模型,以获得卓越的性能.
主要方法:
- 提出了一种新的机器学习模型,集谱预测 (ESP),集成多层感知器 (MLP) 和图形神经网络 (GNN) 模型.
- ESP学习了MLP和GNN光谱预测器输出的最佳权重,以生成查询分子的精细光谱预测.
- 通过将标签混合和多任务在光谱主题分布上的峰值依赖性纳入改进的基线MLP和GNN模型,并根据分子公式分层训练数据.
主要成果:
- 在NIST 2020数据集和PubChem候选集上评估时,ESP模型显示出显著的性能增长,平均排名比MLP基线提高了23.7%,比GNN基线提高了37.2%.
- 对于代谢物注释的最新神经网络方法,ESP实现了性能改进.
- 分析表明,注释性能受训练数据集,候选数据集的大小以及候选分子和目标分子之间的相似性影响.
结论:
- 整体光谱预测 (ESP) 模型在代谢学中的代谢物注释准确性方面取得了重大进展.
- 集成MLP和GNN模型,加上等级学习,为光谱识别提供了更强大的方法.
- 开发的ESP工具,包括代码和训练模型,公开提供,以促进更广泛的采用和该领域的进一步研究.
更多相关视频
11:02Identification and Quantification of Deranged Metabolites in Critically Ill Patients Using NMR-Based Metabolomics
Published on: November 29, 2024
500
14:42Liquid Chromatography Coupled to Refractive Index or Mass Spectrometric Detection for Metabolite Profiling in Lysate-based Cell-free Systems
Published on: September 23, 2021
4.7K
相关概念视频
Mass Spectrum: Interpretation
1.1K
An unknown compound can be established by identifying the molecular ion peak in the mass spectrum. The molecular ion peak is often weak or absent due to the predominance of fragmentation in high-energy electron beams. In such cases, a low-energy electron beam can be used to scan the spectrum to enhance the intensity of the molecular ion peak. Additionally, chemical ionization, field ionization, and desorption ionization spectra are used to obtain a relatively intense molecular ion peak.
To...
To...
1.1K
Mass Spectrum
1.8K
A mass spectrum is the graphical representation of the relative abundance of the charged fragments in an analyte plotted against their mass-to-charge ratio (m/z). The plot's x axis represents the ratio of the mass of the charged fragment to the elementary charge it carries. The y axis of the plot represents the relative abundance of each charged species. The relative abundance is calculated from the signal intensity of each charged species recorded at the detector. The most intense signal...
1.8K
