Molecular Networks and Deep Learning for Targeted HIV Interventions among PWID

点击翻译按钮获取中文摘要

PROJECT SUMMARY Meeting targets set to end AIDS by 2030 requires reaching all populations, particularly those with the highest burden, such as people who inject drugs (PWID). PWID continue to experience some of the most explosive HIV epidemics globally. Injection drug use is increasingly accounting for new HIV infections in both low- and middle-income countries (LMICs) and countries that once saw notable declines in HIV incidence among PWID. Even in countries with notable declines in HIV incidence among PWID, such as the United States, the rise of prescription opioid use has resulted in increased heroin injection, increased overdose rates, and outbreaks of HIV. Combating the HIV epidemic among hard-to-reach populations, such as PWID, requires targeted approaches that consider multiple levels of risk that extend beyond individual-level factors alone. Looking at HIV prevention through the lens of network science can allow us to study and address health disparities on a social and structural level. Limited network studies among PWID have demonstrated that social and spatial networks play a significant role in HIV transmission and may be further leveraged for targeted intervention approaches. However, network data can be challenging to enumerate and analyze, and additional network tools and analytic approaches are needed to take full advantage of the power of social networks for HIV prevention efforts. Given the challenges of collecting data on social connections among PWID, by finding proxies for network data or ways to impute networks we can harness these connections to interrupt HIV transmission. Network-based interventions may not only be more effective at interrupting community transmission than individual-level approaches but could represent the most cost-efficient approach as well – which is crucial given the budgetary and resource constraints programs often face. This study leverages a rare set of longitudinal social and spatial network data along with detailed individual- level data and HIV sequences from over 2,500 PWID in New Delhi, India followed from 2016-21. It aims to explore the use of machine learning and viral phylogenetics as a potential avenue to circumvent network enumeration challenges and produce new analytical strategies to monitor epidemics and model the most effective and resource-efficient intervention approach in a city. In practice, this affords the development of network models that simulate the effect of various network-based intervention strategies on HIV incidence and could be used to inform a wide array of social, behavioral, and pharmacologic interventions. Making network data more accessible can lead to new HIV prevention approaches that guide officials in focusing limited resources for the greatest impact and can provide a greater understanding of the epidemic dynamics.