基于智能计算噬菌体的细菌宿主范围预测

Progress in predicting bacteriophage host ranges by intelligent computing

摘要:

针对噬菌体的细菌宿主范围预测对于深入理解噬菌体及将其作为抗生素替代用于生物疗法具有重要意义。传统生物实验方法确定噬菌体的细菌宿主范围受到极有限的噬菌体可培养性和严苛的培养条件限制，而高通量测序技术所提供的海量基因组或宏基因组序列提供了噬菌体及细菌重要的序列信息，因此智能计算为预测噬菌体的细菌宿主范围提供了可行方法。本文从智能计算的角度对噬菌体的细菌宿主范围预测研究进行系统梳理，从噬菌体感染细菌的过程入手，描述配对预测模型所依赖的特征及其生物合理性，归纳宿主范围预测的智能模型、建模原理及预测策略，总结建模训练和评估所依赖的参考数据集与真实数据及评价指标。本文特别注意挖掘和分析各信息手段、模型、方法其背后的生物合理性及其依赖的生物机理。本综述期望推动基于智能算法的噬菌体的细菌宿主范围预测研究发展，并探索将生物先验结合人工智能实现噬菌体侵袭细菌宿主的本质机理推断，同时也为基于噬菌体的临床应用提供参考与借鉴。

关键词:微生物组;噬菌体-宿主相互作用;预测模型;智能算法;机器学习;神经网络

Abstract:

The prediction of bacteriophage host ranges is of great significance for the basic research and clinical application of bacteriophages. The conventional biological experimental methods are limited by the poor culturability of bacteriophages and strict cultivation conditions. The availability of massive genome or metagenome sequencing data provides the sequence signature of bacteriophages and bacteria. Therefore, intelligent computing serves as a feasible way to predict bacteriophage host ranges. This paper systematically reviews the studies about intelligent computing-based prediction of bacteriophage host ranges. Starting from the process of bacteriophage infecting bacteria, we describe the feature source and biological rationality of prediction models, analyze the typical intelligent models and their prediction principles, and list all the reference datasets, real-world datasets, and evaluation indicators. The review aims to improve the understanding on the mechanism of bacteriophages in invading bacterial hosts and promote the usage of bacteriophages as antibiotic substitutes in biological therapy.

Key words:microbiome;bacteriophage-host interactions;prediction model;intelligent computing;machine learning;neural network