Applications of Machine Learning for the Classification of Porcine Reproductive and Respiratory Syndrome Virus Sublineages Using Amino Acid Scores of ORF5 Gene

Jeong‐Hoon Kim, Kyuyoung Lee, Ruwini Rupasinghe, Shahbaz Rezaei, Beatriz Martínez‐López, Xin LiuFrontiers in Veterinary Science 2021 · DOI

TL;DR

This paper introduces a machine learning approach to rapidly and accurately classify strains of the Porcine Reproductive and Respiratory Syndrome Virus (PRRSV) using amino acid scores from the ORF5 gene. By leveraging four ML algorithms, the study achieves near-perfect classification accuracy with significantly reduced computational time compared to traditional phylogenetic methods, offering a promising tool for real-time surveillance and control strategies.

What Problem Does This Paper Address?

Porcine Reproductive and Respiratory Syndrome Virus (PRRSV) is a major economic threat to the global swine industry, causing reproductive failures and respiratory disease. Effective control and prevention strategies rely heavily on accurately classifying different PRRSV strains. Historically, methods like Restriction Fragment Length Polymorphism (RFLP) have been used, but they suffer from unstable accuracy due to the virus's genetic variability. Phylogenetic analysis offers better accuracy but becomes computationally prohibitive as the number of viral sequences grows. This paper addresses the need for a classification method that is both accurate and computationally efficient, capable of handling large datasets in real-time or near real-time.

Methods

The study employs four machine learning (ML) algorithms for classification: Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Multilayer Perceptron (MLP).

The core innovation lies in how the genetic data is transformed:

  • Amino Acid Scoring: Instead of directly using DNA or RNA sequences, the amino acid sequences of the ORF5 gene are converted into numerical scores. Each amino acid is represented by a vector of five scores reflecting its physicochemical properties (polarity, secondary structure, molecular volume, codon diversity, and electrostatic charge). This transformation is crucial because raw genetic sequences are long strings of letters, making direct ML application challenging without significant information loss.
  • Feature Engineering: The 200 amino acids in the ORF5 gene, each represented by 5 scores, result in a 1000-dimensional feature vector for each PRRSV strain.
  • Classification and Evaluation: The ML algorithms are trained on these feature vectors to classify strains into four predefined clades (Lineage 5 and three sub-lineages within Lineage 1). Performance is evaluated using Area Under the Curve (AUC) of the Receiver Operating Characteristic (ROC) curve, precision, recall, and F1-score, alongside time consumption.

Datasets

The study utilizes a dataset of 1931 field PRRSV strains collected in the US between 2012 and 2020. The ORF5 gene sequences from these strains were used. Phylogenetic analysis, based on these sequences, was employed to initially label the strains into one of four clades: Lineage 5 (L5) or three clades within Lineage 1 (L1A, L1B, L1C). No new benchmark datasets were introduced; the study uses existing field data.

Key Insights and Ideas

  • Amino Acid Scores as a Powerful Feature: Transforming amino acid sequences into physicochemical scores is a viable and effective strategy for ML-based classification of viral genetic data, overcoming the limitations of direct sequence analysis.
  • ML Outperforms Traditional Methods: The ML approaches demonstrate superior accuracy and significantly faster processing times compared to traditional methods like RFLP and even phylogenetic analysis for large datasets.
  • Key Amino Acid Positions Identified: Random Forest analysis successfully pinpointed specific amino acid positions (notably the 26th and 170th) within the ORF5 gene that are critical for differentiating PRRSV clades. This suggests these positions may be linked to important biological characteristics or evolutionary divergence.
  • Efficiency for Large-Scale Surveillance: The ability of ML models to classify thousands of sequences in seconds makes them ideal for real-time or near real-time epidemiological surveillance and rapid decision-making in disease control.

Key Results

  • High Accuracy: All four ML algorithms achieved very high classification accuracy, with AUC values consistently above 0.99 when using a sufficient number of amino acid positions (2 or more). Even with just the 26th position, accuracy was around 80%.
  • Rapid Classification: The ML models classified the entire dataset of 1931 strains in under 2.5 seconds, a dramatic improvement over computationally intensive phylogenetic methods.
  • Key Positions: The 26th amino acid position in the ORF5 gene was identified as having the highest importance score for classification, showing significant differences between L5 and L1 clades. Including the 170th position further improved classification accuracy for all ML approaches.
  • RFLP Limitations: While RFLP could classify strains into L5 and L1 clades, it struggled with distinguishing between the sub-lineages within L1, and some RFLP types were ambiguous across multiple clades.

Limitations and Future Work

  • Genetic Homogeneity: The authors acknowledge that strong genetic homogeneity within clades, compared to heterogeneity among clades, might have inflated the observed accuracy.
  • Complexity of Multi-Class Classification: The paper notes that ML approaches might show lower accuracy and longer processing times for more complex multi-class classification problems with highly complex genomic data.
  • Future Directions: The authors suggest exploring more complex ML approaches for classification using larger genetic information. They also propose investigating ML for predicting multiple phenotypic and antigenic characteristics and identifying key substitutions related to unique biological characteristics of field PRRSV strains. This could lead to more sophisticated vaccine lineage selection based on genetic relatedness.

Summary updated

Abstract

Porcine reproductive and respiratory syndrome is an infectious disease of pigs caused by PRRS virus (PRRSV). A modified live-attenuated vaccine has been widely used to control the spread of PRRSV and the classification of field strains is a key for a successful control and prevention. Restriction fragment length polymorphism targeting the Open reading frame 5 (ORF5) genes is widely used to classify PRRSV strains but showed unstable accuracy. Phylogenetic analysis is a powerful tool for PRRSV classification with consistent accuracy but it demands large computational power as the number of sequences gets increased. Our study aimed to apply four machine learning (ML) algorithms, random forest, k-nearest neighbor, support vector machine and multilayer perceptron, to classify field PRRSV strains into four clades using amino acid scores based on ORF5 gene sequence. Our study used amino acid sequences of ORF5 gene in 1931 field PRRSV strains collected in the US from 2012 to 2020. Phylogenetic analysis was used to labels field PRRSV strains into one of four clades: Lineage 5 or three clades in Linage 1. We measured accuracy and time consumption of classification using four ML approaches by different size of gene sequences. We found that all four ML algorithms classify a large number of field strains in a very short time (<2.5 s) with very high accuracy (>0.99 Area under curve of the Receiver of operating characteristics curve). Furthermore, the random forest approach detects a total of 4 key amino acid positions for the classification of field PRRSV strains into four clades. Our finding will provide an insightful idea to develop a rapid and accurate classification model using genetic information, which also enables us to handle large genome datasets in real time or semi-real time for data-driven decision-making and more timely surveillance.
BibTeX
@inproceedings{kim2021applications,
  author = {Jeong‐Hoon Kim and Kyuyoung Lee and Ruwini Rupasinghe and Shahbaz Rezaei and Beatriz Martínez‐López and Xin Liu},
  title = {{Applications of Machine Learning for the Classification of Porcine Reproductive and Respiratory Syndrome Virus Sublineages Using Amino Acid Scores of ORF5 Gene}},
  booktitle = {Frontiers in Veterinary Science},
  year = {2021},
  doi = {10.3389/fvets.2021.683134},
}
Figure