ProTAMAR用于识别蛋白质序列的扭转角
ProTAMAR for identifying the torsion angle of protein sequences
姜博文1
作者信息
- 1. 浙江理工大学计算机科学与技术学院,浙江 杭州 310018
- 折叠
摘要
蛋白质的扭转角控制着蛋白质的空间构象和功能.为了提升蛋白质序列的扭转角预测性能,本文提出一种新的深度学习模型ProTAMAR.在传统蛋白质序列编码和多序列对比结果的基础上,通过引入蛋白质预训练编码以捕获高维特征表示,设计多头注意力机制和扩张卷积模块用于提取全局序列信息和局部上下文信息.在蛋白质基准数据集中广泛测试,ProTAMAR模型优异.通过实验证实本文设计的预训练特征和引入的网络框架为蛋白质序列扭转角预测任务提供了更具价值的生物学线索和更高效的提取方式.
Abstract
The torsion angle of proteins controls the spatial conformation and function of proteins.To improve the performance of torsion angle prediction of protein sequences,a new deep learning model,ProTAMAR,is proposed.Based on the traditional protein sequence coding and multiple sequence comparison results,a protein pre-training coding is introduced to capture high-dimensional feature representation,and a multi-headed attention mechanism and a dilated convolution module are designed for extracting global sequence information and local contextual information.The ProTAMAR model is tested extensively in protein benchmark datasets with excellent results.It is experimentally confirmed that the pre-trained features designed and the network framework introduced in this paper provide more valuable biological cues and more efficient extraction for protein sequence torsion angle prediction tasks.
关键词
蛋白质/扭转角/多头自注意力机制/扩张卷积/ProteinBERTKey words
protein/torsion angle/multi-headed self attentive mechanism/dilated convolution/ProteinBERT引用本文复制引用
出版年
2023
NETL
NSTL
维普
万方数据