[Paper Review] AI4AI: Quantitative Methods for Classifying Host Species from Avian Influenza DNA Sequence

Woo Yong Choi, Kyu Ye Song+1 more|arXiv (Cornell University)|Feb 26, 2018

Influenza Virus Research Studies7 references3 citations

TL;DR

This study proposes machine learning and deep learning models to classify host species from avian influenza DNA sequences, using raw genetic data to predict species with probabilistic outputs. The best deep learning models achieve 47% top-1 accuracy and 82% top-3 accuracy across 11 host species classes.

ABSTRACT

Avian Influenza breakouts cause millions of dollars in damage each year globally, especially in Asian countries such as China and South Korea. The impact magnitude of a breakout directly correlates to time required to fully understand the influenza virus, particularly the interspecies pathogenicity. The procedure requires laboratory tests that require resources typically lacking in a breakout emergency. In this study, we propose new quantitative methods utilizing machine learning and deep learning to correctly classify host species given raw DNA sequence data of the influenza virus, and provide probabilities for each classification. The best deep learning models achieve top-1 classification accuracy of 47%, and top-3 classification accuracy of 82%, on a dataset of 11 host species classes.

Motivation & Objective

• To develop quantitative methods for classifying host species of avian influenza using raw DNA sequence data.
• To reduce reliance on time-consuming laboratory testing during outbreak emergencies by enabling rapid, in silico classification.
• To provide probabilistic classification outputs for improved diagnostic confidence and decision-making.
• To evaluate the performance of machine learning and deep learning models on a diverse dataset of 11 host species.
• To support early detection and response to avian influenza outbreaks by accelerating host species identification.

Proposed method

• The study employs machine learning and deep learning models trained on raw DNA sequences of avian influenza viruses.
• Feature representation is derived directly from nucleotide sequences without manual feature engineering.
• Convolutional neural networks (CNNs) and other deep learning architectures are applied to capture sequence patterns predictive of host species.
• Models are evaluated using top-1 and top-3 classification accuracy metrics on a dataset of 11 host species.
• Probabilistic outputs are generated for each host species prediction, enabling uncertainty assessment.
• The approach leverages transferable sequence representations to improve generalization across host species.

Experimental results

Research questions

• RQ1 Can machine learning models accurately classify host species from raw avian influenza DNA sequences without prior biological feature engineering?
• RQ2 What is the performance of deep learning models in predicting host species across a diverse set of 11 avian and mammalian hosts?
• RQ3 How do top-1 and top-3 classification accuracies compare across different model architectures?
• RQ4 Can probabilistic outputs from the models improve confidence in host species predictions during outbreak scenarios?
• RQ5 To what extent can these models reduce the need for time-intensive laboratory testing in emergency response?

Key findings

• The best deep learning models achieve a top-1 classification accuracy of 47% on a dataset of 11 host species.
• The top-3 classification accuracy reaches 82%, indicating strong performance when considering the most likely three predictions.
• The models successfully classify host species using only raw DNA sequence data, eliminating the need for complex preprocessing or biological priors.
• Probabilistic outputs are generated for each prediction, supporting risk assessment and decision-making in outbreak settings.
• The results demonstrate the feasibility of using AI to accelerate host species identification during avian influenza outbreaks.
• The study establishes a foundation for scalable, low-resource diagnostic tools applicable in regions with limited laboratory infrastructure.