Development of a joint machine learning/de novo assembly system for resolving viral quasispecies

项目总结 乙型病毒性肝炎(乙肝)在全球2.5亿人中造成慢性感染；慢性病是 在上升中，大约三分之一的世界人口(20亿)有血清学证据表明 曝光。乙肝病毒与丙型肝炎病毒和艾滋病病毒的混合感染是药物使用障碍的一个隐藏后果 流行病。病毒种群具有极高的序列多样性和快速进化，这解释了 疫苗失败率和对现有疗法的病毒耐药性，并使发现持久的疗法 极具挑战性。下一代测序(NGS)是评估主机内的方法的选择 病毒种群，被称为“准种”。虽然获取了一大组短DNA测序读数， 代表准物种中的病毒粒子，计算技术在分析能力上是有限的， 导致复杂的乙肝病毒基因组结构的分辨率特别低。另一个挑战是组装 NGS读取代表宿主基因组短片段的完整菌株(单倍型)，而不知道 它们在样本中的真实存在。为了应对这些挑战，GATACA正在开发针对病原体的 支持治疗的生物信息学软件GAT-ML(GATACA组装工具-机器学习[ML]) 发现并改进感染控制。其专门设计的算法采用了新颖的ML方法 调整和修改以帮助基因组组装，使GAT-ML能够重建完整的病毒 通过学习序列的“语言”，单倍型和种群。最初是为乙肝样本量身定做的， GAT及其新的ML系统将在第一阶段与以下具体系统集成进行可行性测试 目标： 1.明确目标1.构建联合学习体系。培训和测试自然语言处理(NLP)方法 乙肝病毒基因变异。 2.具体目标2.在GAT(GAT-ML)中实现和测试机器学习方法。 我们期待着一种用于表征乙肝病毒单倍型的有效工具，并通过多源数据集进行验证，以及 对线下和集成方法进行广泛的测试和标杆。

PROJECT SUMMARY Viral hepatitis from hepatitis B (HBV) establishes chronic infections in >250M people worldwide; chronicity is on the rise, and approximately one-third of the world’s population (2 billion) has serologic evidence of exposure. HBV coinfection with HCV and HIV is a hidden consequence of the substance use disorder epidemic. Viral populations have extremely high sequence diversity and rapidly evolve, which explains the vaccine failure rates and viral resistance to existing therapies and makes discovering lasting therapies extremely challenging. Next Generation Sequencing (NGS) is the method of choice to assess the intra-host virus population, termed a “quasispecies”. While a large set of short DNA sequencing reads are acquired that represent the virions in the quasispecies, computational technologies are limited in their analysis capabilities, resulting in particularly low resolution of complex HBV genomic structures. Another challenge is assembling NGS reads representing short fragment of the host genome into full strains (haplotypes) without knowledge of their true occurrence in the samples. To meet these challenges, GATACA is developing pathogen-specific bioinformatics software, GAT-ML (GATACA Assembly Tool – machine learning [ML]) to support treatment discovery and improve infection control. Its specifically designed algorithm utilizes novel ML methodologies adapted and modified for assisting genome assembly that will allow GAT-ML to reconstruct complete viral haplotypes and populations by learning the ‘language’ of the sequences. Tailored initially for HBV samples, GAT and its new ML system will be integrated for feasibility testing in this Phase I with the following Specific Aims: 1. Specific Aim 1. Build a joint learning system. Train and test natural language processing (NLP) methods on HBV genetic variation. 2. Specific Aim 2. Implement and test the machine learning methods in GAT (GAT-ML). We anticipate a working tool for characterizing HBV haplotypes, validated with multi-sourced datasets, and extensive testing and benchmarking of offline and integrated methods.