Dynamic stochastic block modelling for analysing recombination in HIV

艾滋病病毒在世界范围内仍然是一个巨大的负担，每年有170万新感染者（联合国艾滋病规划署，2019年）。抗逆转录病毒疗法（ART）的推广已经减少了艾滋病相关死亡和传播的数量，但为了遏制进一步的感染，联合国艾滋病规划署的目标是95%的人口应该知道自己的状况，95%的人应该接受治疗，95%的人应该受到病毒的抑制。确定新感染的驱动因素将有助于确定有待弥补的差距。比较不同患者的病毒序列可用于流行病学研究。聚合酶基因（pol）的HIV序列数据是常规收集用于耐药性测试的，但随后可以用于这些目的，一旦匿名，只保留基本的人口统计信息。遗传距离（即任意两种病毒之间突变差异的数量）可以用来将密切相关的病毒联系在一起。（较低的遗传距离表明他们有一个共同的祖先）。每个复制周期都会将HIV突变引入基因组。据说突变有它自己的“时钟”，所以随着时间的推移，变化以一种可预测的方式平均积累起来。因此，遗传距离和采样时间，可以用来绘制链接，推断网络，传播模式，以及网络的其他特征，如程度分布。这些与人口统计信息相结合的见解可以为公共卫生政策提供信息。例如，来自高危人群的个人可能被建议采取暴露前预防（PrEP）。艾滋病病毒的多样性非常高，因为这种病毒在人类身上进化了大约100年，远远早于它被首次描述出来。它被划分为主要的谱系（亚型），形成于它扩张的早期。当一个人感染了一种以上的艾滋病毒变体时，两者之间可能发生重组，产生混合病毒，从而产生更多的多样性。这几乎肯定会发生在来自同一感染的两种相同病毒之间，但由于新病毒与双亲相同，因此无法检测到。当高度分化的病毒重组时（例如来自不同亚型的病毒），这种情况变得更加明显，因为有足够的信号来区分两种亲本病毒。不同病毒之间的这种重组过程打破了联系，其中一半的基因组可能与第一个亲本病毒相连，另一半与第二个亲本病毒相连。现在，如果要在连锁分析中考虑整个序列，则不会建立任何联系，因为新的序列现在与父母双方完全不同。当艾滋病毒沿着传播网络移动时，它偶尔会发现自己是双重感染的一部分，并可能参与重组事件。这可能在任何时间点发生，随着其他突变的增加，分子钟使病毒向前移动，这使得发现它变得更加困难。动态随机块建模是一种对网络数据建模的方法，在我们的案例中，它将被用来寻找一段时间内相似病毒的群体或社区。随着时间的推移，这种方法将更好地对艾滋病毒多样性进行分类并建立网络模型；非常适合于快速进化的重组病毒。将进行仿真实验来测试该原理并验证该方法。最后，我们将把它应用于来自乌干达的接近全基因组的HIV数据。本研究将由加拿大安大略省西安大略大学病理及检验医学系潘嘉文副教授指导。

MRC : Heather Grant : MR/N013166/1 HIV is still a huge burden world-wide, with 1.7 million new infections each year (UNAIDS, 2019). The roll out of anti-retroviral therapies (ART) has worked to reduce the numbers of AIDS related deaths and onward transmissions, but to curb further infections still, UNAIDS goals are that 95% of the population should know their status, 95% of those should be on treatment, and 95% of those should be virally supressed. Characterising drivers of new infections will help to identify gaps to be closed. Comparing viral sequences from different patients can be used for epidemiological studies. HIV sequence data for the polymerase gene (pol) is routinely collected for drug-resistance testing, but can then be used secondarily for these purposes, once anonymized, keeping only basic demographic information. Genetic distance (that is, the number of mutational differences between any two viruses) can be used to link closely related viruses together. (A lower genetic distance suggests they shared a common ancestor more recently). HIV mutations are introduced into the genome with each replication cycle. Mutation is said to have its own 'clock' so that changes builds up, on average, in a predictable way over time. Therefore, the genetic distance and time of sampling, can be used to draw linkage, infer networks, patterns of transmission, and other characterisations of the network such as degree distribution. These insights tied with demographic information can inform public health policy. For instance, individuals from groups deemed at high-risk might be advised to take pre-exposure prophylaxis (PrEP). HIV diversity is extremely high, since the virus has been evolving in humans for maybe a hundred years, long before it was first described. It is classified into major lineages (subtypes) that formed early on during its expansion. Where an individual is infected with more than one HIV variant, recombination between the two can occur, creating a hybrid virus, and thus more diversity. This almost certainly happen between two identical viruses from the same infection, but will be undetectable since the new virus is the same as both parents. Where highly divergent viruses recombine, (such as those from different subtypes), this becomes more obvious as there is enough signal to distinguish the two parental viruses. This process of recombination between divergent viruses breaks apart linkages, where one half of the genome might link to the first parental virus, and the other half to the second. Now, if the whole sequence was to be considered in a linkage analysis, no connections would be made as the new sequence is now sufficiently different to both parents. As HIV moves along the transmission network, it will occasionally find itself part of a dual infection, and may take part in a recombination event. This could happen at any time point in time, making it more difficult to spot, as other mutations build up, and the molecular clock moves the virus forward. Dynamic Stochastic Block Modelling is a way of modelling network data, and in our case will be used to find groups or communities of similar viruses over time. This approach will better classify HIV diversity and model networks over time; highly appropriate for a fast-evolving recombinogenic virus. Simulation experiments will be carried out to test the principle and validate the approach. Finally, we will apply this to near-full genome HIV data from Uganda. This research will be undertaken under the supervision of Associate Professor Art Poon in the Department of Pathology and Laboratory Medicine at Western University, Ontario, Canada.