QRIS: a machine learning framework to investigate the determinants of retroviral integration specificity

Abstract

<p indent="0mm">Retroviral activity is linked with various diseases. Retrovirus can integrate a copy of its genomic DNA into the host genome and store it as the provirus. Thus, it threatens the host cell by interrupting the host genome architecture and transcribing its provirus for detrimental expansion. However, depending on where the retrovirus integrates, the host has corresponding defense mechanisms to repress the transcription of the provirus and eliminate harm. Therefore, the integration site selection is a vital process for retrovirus’ fate. Many efforts have been made to understand its integration specificity and find diverse DNA motif preferences across retrovirus genera. However, the effect of genome-wide DNA structural properties, such as DNA shapes, on retroviral integration is unclear. Therefore, we systematically investigated this issue on six retroviruses that are representative of four genera. We devised QRIS (quantify the retrovirus integration specificity), a machine learning framework to assess the effect of DNA shape on large-scale retroviral integration sites. We observed that the DNA shape could independently or cooperatively work with the DNA motif to regulate retroviral integration. Based on this, we classified these retroviruses into three categories: StrongFavor retrovirus, WeakFavor retrovirus, and Strongshape retrovirus. Interestingly, the Strongshape retrovirus can gain specificity through DNA shapes without a particular DNA motif. Our qualitative and quantitative evaluation of DNA shape and motif revealed their diverse roles in regulating the retroviral integration specificity. Our findings may help control the lentivirus vector more precisely for gene therapy and disturb retroviral integration during pathogenesis.