Speaker
Description
HIV-1 has a high rate of recombination that has significantly shaped its evolutionary history. Identifying these past recombination events will require accurate and scalable computational methods. We previously adapted the dynamic stochastic block model (DynSBM) from social network analysis to recombination detection, demonstrating greater accuracy on simulated data. We describe further improvements on DynSBM by applying an expanded dataset of HIV-1 genomes, including early sequences. We present a novel Bayesian implementation of DynSBM tailored for recombination analysis.
We partitioned an alignment of $n=718$ HIV-1 sequences into 73 windows, and inferred a graph for each window from the TN93 distance matrix. DynSBM identified $K=30$ communities from the resulting series of graphs. We assume a transition between communities as we traverse a sequence represents a putative recombination breakpoint. Transitions were removed if they reverted to the previous community after one window, retaining 65% inferred breakpoints. Based on the inferred transition rate matrix $Q$, we applied an elbow method to $Q^{-1}$ to cluster the communities into nine proposed subtypes, which were highly concordant with known subtypes (normalized mutual information = 0.90). Because DynSBM requires complete data, we incorporated partial genomes, including sequences from historical samples (1959-1978, $n=9$) by assigning communities from the $k$-nearest neighbours (TN93) for each window. We extracted putative non-recombinant intervals for sets of three windows from the alignment by removing sequences with any transitions. Intervals retained about 39%-62% of sequences, with substantial turnover among intervals.
Key limitations of DynSBM are that it is not designed for recombination detection and cannot quantify uncertainty. To address this, we have implemented a custom Bayesian version of DynSBM in R (bayblocks). With data represented as a sequence of adjacency matrices, we use the Gibbs algorithm to sample community memberships and transition rates, incorporating biologically realistic constraints.
| Expedited Notification | No thanks, I do not require Expedited Notification |
|---|
