Machine learning methods for predicting human-adaptive influenza A virus reassortment based on intersegment constraint.

Abstract

Introduction: It is not clear about mechanisms underlining the inter-segment reassortment of Influenza A viruses (IAVs).We analyzed the viral nucleotide composition (NC) in coding sequences,examined the intersegment NC correlation, and predicted the IAV reassortment using machine learning (ML) approaches based on viral NC features.

Methods: Unsupervised ML methods were used to examine the NC difference between human-adapted and zoonotic IAVs. Supervised ML models of random forest classifier (rfc) and multiple-layer preceptor (mlp) were developed to predict the human adaption to IAVs.

Results: Our results demonstrated that the frequencies of thymine, cytosine, adenine,and guanine (t, c, a, and g), as well as the content of gc/at were consistently high or low for the segments of PB2, PB1, PA, NP, M1, and NS1 (ribonucleoprotein plus [RNPplus]), between mammalian and avian IAVs or between influenza B viruses (IBVs) and IAVs.RNPplus NC negatively correlated with the NC for HA, NA, and M1 (envelope protein plus [EPplus]). The human-adapted NC accurately discriminated between human IAVs and avian IAVs. A total of 221,184 simulated IAVs with pd09H1N1 EPplus and with RNPplus from other IAV subtypes indicated a high adaption of the RNPplus, from H6N6, H13N2, and H13N8 and other IAVs.

Discussion: In summary, there is a distinct human adaption-specific genomic NC between human IAVs and avian IAVs. The intersegment NC correlation constrains segment reassortment. This study presents a novel strategy for predicting IAV reassortment based on viral genetic compatibility.