Estimating the timing of HIV infection from unmutated sequences

For HIV, the time since infection can be estimated from sequence data for acutely infected samples. One popular approach relies on the star-like nature of phylogenies generated under exponential population growth, and the resulting Poisson distribution of mutations away from the founding variant. However, real-world complications, such as APOBEC hypermutation and multiple-founder transmission, present a challenge to this approach, requiring data curation to remove these signals before reasonable timing estimates may be obtained. Here we suggest a simple alternative approach that derives the timing estimate not from the entire mutational spectrum but from the proportion of sequences that have no mutations. This can be approximated quickly and is robust to phenomena such as multiple founder transmission and APOBEC hypermutation. Our approach is Bayesian, and we adopt a conjugate prior to obtain closed form posterior distributions at negligible computational expense. Using real data and simulations, we show that this approach provides accurate timing estimates and credible intervals without the inconvenience of data curation and is robust to complicating phenomena that can mislead existing approaches or cause them to fail entirely. For immediate use we provide an implementation via Google Sheets, which offers bulk analysis of multiple datasets, as well as more detailed individual-donor analyses. For inclusion in data processing pipelines we provide implementations in three languages: Julia, R, and Python.