2026-06-14T03:51:07Zhttps://repisalud.isciii.es/rest/oai/request

oai:repisalud.isciii.es:20.500.12105/167792024-09-27T19:13:31Zcom_20.500.12105_2052com_20.500.12105_2051col_20.500.12105_19609

00925njm 22002777a 4500 dc Guerrero-Tamayo, Ana author Sanz Urquijo, Borja author Casado, Concepcion author Moragues, María Dolores author Olivares, Isabel author Pastor-López, Iker author 2023 In this article, we introduce a novel methodology for characterizing viral genetic features: the Unified Methodology of recombinant virus Identification (UMI). Our methodology converts genomic sequences into spectrograms, applies transfer learning using a pre-trained Convolutional Neural Network (CNN), and employs interpretability tools to identify the genomic regions relevant for characterizing a viral sequence as recombinant. The UMI methodology does not necessitate multiple sequence alignment or manual adjustments. As a result, it operates much faster, has low computational demands, and is capable of handling substantial amounts of data. To validate this, we applied UMI to one extensively studied and documented case: HIV-1 genetic recombination. We worked with all identified HIV-1 complete sequences (13554 sequences up to 2020), searching for mathematical patterns, signatures, that characterize an HIV-1 sequence as recombinant. CNN’s hit rate (test accuracy) is 94%, with consistent and differentiated decision areas in each category. Using interpretability tools, we verified that the hot zones were similar for sequences of the same subtype and phylogenetic proximity. The leading areas for classifying a sequence as recombinant or non-recombinant are coincident with genomic regions that play a key role in genetic recombination processes. By applying UMI methodology we found that there is indeed a genome mathematical pattern that assesses an HIV-1 sequence as recombinant. In addition, we located its position. Considering expert knowledge, our results showed a substantial, robust and biologically-consistent hit rate. This type of solution can successfully guide the location and subsequent characterization of relevant areas, avoiding the heavy analysis of multiple sequence alignment and manual adjustments. IEEE ACCESS. 2023,11:95796-95812. 10.1109/ACCESS.2023.3311752 2169-3536 IEEE Access http://hdl.handle.net/20.500.12105/16779 Convolutional neural network Deep learning Genetic recombination Genome mathematical pattern Genome mathematical signature HIV-1 Discovering Mathematical Patterns Behind HIV-1 Genetic Recombination: A New Methodology to Identify Viral Features