Project Summary An important global public health priority is to develop new methods for identifying populations at greatest HIV risk, understanding HIV transmission network patterns, and intervening to reduce network risk. HIV testing is important to effect positive sexual behavior changes, and is an entry point to treatment, care, and psychosocial support. At the end of 2016, an estimated 1.1 million persons aged 13 and older were living with HIV infection in the United States, including an estimated 162,500 (14%) persons whose infections had not been diagnosed. In addition, many persons with HIV are tested late in the course of infection. Late testing results in missed opportunities for prevention and treatment of HIV, and increased risk for transmission to their partners. Current status ? A number of epidemiological studies have employed social network theory/concepts and applied network analytical techniques to examine the structural characteristics of HIV transmission networks through phylogenetic link (HIV-1 pol sequences) and/or sexual/social/drug-using contacts among MSM. These studies, however, usually reduce the network information to summary information, consider only a subset of network variables, and/or use one layer of multi-dimension networks determining transmission paths such as only the social, sexual, contact, and venue perspectives. Challenges: The complexity of data that is important for HIV infection risk analysis makes it challenging to conduct risk and transmission prediction. More specifically, we are facing two challenges: (1) How to develop a mechanism to faithfully and flexibly represent the multi-dimensional network data collected from different sources at different time periods; (2) Once the data has been integrated, how to fully leverage the data to develop a risk prediction algorithm that considers the multi-dimensional networks with substantially interrelated factors in a comprehensive manner. Goals - We hypothesize that deep learning-based informatics approaches can provide a novel way for HIV infection risk prediction. In close collaboration with the public health department, we will construct a comprehensive framework that combines population-based molecular, behavior, and contact/partner tracing information including venue affiliation data and individual sex and drug-using behaviors, as well as existing locally collected cohort data. Using this dynamically collected data we will then develop practical deep-learning algorithms that leverage the comprehensive framework for cluster growth and identifying newly infected population. This proposal focuses specifically on ongoing epidemic growth among populations most at risk, including young men who have sex with men (MSM), which remain highly vulnerable to HIV in the U.S.

R56AI150272

2020-09-01

2021-08-31