Seq2Hosts

Seq2Hosts predicts potential hosts of coronaviruses from spike gene mononucleotide and dinucleotide biases using support vector machine (SVM) and Mahalanobis distance discriminant (MD) models to inform interspecies transmission studies.

Key Features:

• Dual-model approach: Uses complementary SVM and Mahalanobis distance discriminant (MD) models for host classification.
• Nucleotide-bias feature set: Analyzes nineteen parameters derived from spike gene sequences that capture mononucleotide and dinucleotide biases.
• Cross-validation performance: Achieved 99.86% accuracy for the SVM model and 98.08% accuracy for the MD model in leave-one-out cross-validation on a dataset of 730 representative coronaviruses.
• External validation: Accurately predicted hosts for an additional set of 47 coronaviruses consistent with other researchers' conclusions or speculations.

Scientific Applications:

• Interspecies transmission studies: Supports analysis of host-specific nucleotide signatures to investigate coronavirus host range and transmission routes.
• Virology and epidemiology: Aids assessment of the threat posed by emerging human pathogens such as SARS-CoV and MERS-CoV by predicting likely reservoir or intermediate hosts.
• Zoonotic risk identification: Facilitates preemptive identification of species that could harbor zoonotic coronaviruses to inform surveillance and intervention strategies.
• Public health support: Provides data to inform early intervention strategies and public health policy regarding potential animal reservoirs of coronaviruses.

Methodology:

Compute nineteen spike-gene-derived parameters representing mononucleotide and dinucleotide biases, apply SVM and Mahalanobis distance discriminant (MD) models for host classification, evaluate performance by leave-one-out cross-validation on 730 representative coronaviruses, and validate predictions on an additional set of 47 coronaviruses.