Background: While it has been suggested that host microRNAs (miRNAs) may downregulate viral gene expression as an antiviral defense mechanism, such a mechanism has not been explored in the influenza virus for human flu studies. As it is difficult to conduct related experiments on humans, computational studies can provide some insight. Although many computational tools have been designed for miRNA target prediction, there is a need for cross-species prediction, especially for predicting viral targets of human miRNAs. However, finding putative human miRNAs targeting influenza virus genome is still challenging.

Results: We developed machine-learning features and conducted comprehensive data training for predicting interactions between H1N1 genome segments and host miRNA. We defined our seed region as the first ten nucleotides from the 5' end of the miRNA to the 3' end of the miRNA and integrated various features including the number of consecutive matching bases in the seed region of 10 bases, a triplet feature in seed regions, thermodynamic energy, penalty of bulges and wobbles at binding sites, and the secondary structure of viral RNA for the prediction.

Conclusions: Compared to general predictive models, our model fully takes into account the conservation patterns and features of viral RNA secondary structures, and greatly improves the prediction accuracy. Our model identified some key miRNAs including hsa-miR-489, hsa-miR-325, hsa-miR-876-3p and hsa-miR-2117, which target HA, PB2, MP and NS of H1N1, respectively. Our study provided an interesting hypothesis concerning the miRNA-based antiviral defense mechanism against influenza virus in human, i.e., the binding between human miRNA and viral RNAs may not result in gene silencing but rather may block the viral RNA replication.