Predicting Antigenic Distance from Genetic Data for PRRSV-Type 1: Applications of Machine Learning

Makau, Dennis N.; Prieto, C.; Martínez-Lobo, Francisco Javier; Paploski, Igor Adolfo Dexheimer; VanderWaal, Kimberly

doi:10.1128/spectrum.04085-22

Cited by 4 publications

(3 citation statements)

References 89 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While the ORF5 gene is immunologically important, other parts of the genome contribute to the antigenicity and virulence (20, 44, 45). Whole genome data would be needed to understand the interplay between genotype and phenotype, and observed clinical manifestations are also influenced by external factors (e.g., co-infections); science has not yet progressed to the point that we can predict phenotype from whole genomes for PRRSV (22). That being said, sequencing conducted by animal health professionals is often conducted for epidemiological monitoring purposes, and this informed the level of granularity that we tried to achieve in this analysis.…”

Section: Discussionmentioning

confidence: 99%

“…Within the ∼15 kb PRRSV-2 genome, open reading frame 5 (ORF5) encodes for a major envelope protein (glycoprotein 5 - GP5), which is involved in inducing virus neutralizing antibodies and cross-protection among PRRSV variants (18–20). Even though ORF5 accounts for only 4% of the genome, its genetic variability and apparent immunologic importance (3, 20–22) has made this gene the target of nearly all genetic sequencing conducted by the swine industry, with thousands of sequences generated per year in the U.S. alone (4). Stakeholder preference for ORF5 rather than whole genome sequencing also relates to lower cost, rapid turnaround time, and the higher probability of successfully obtaining a sequence from samples of various types and quality.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Phylogenetic-based methods for fine-scale classification of PRRSV-2 ORF5 sequences: a comparison of their robustness and reproducibility

VanderWaal,

Pamornchainvakul,

Kikuti

et al. 2024

Preprint

Self Cite

View full text Add to dashboard Cite

Disease management and epidemiological investigations of porcine reproductive and respiratory syndrome virus-type 2 (PRRSV-2) often rely on grouping together highly related sequences. In the USA, the last five years have seen a major paradigm shift within the swine industry when classifying PRRSV-2, beginning to move away from RFLP (restriction fragment length polymorphisms)-typing and adopting the use of phylogenetic lineage-based classification. However, lineages and sub-lineages are large and genetically diverse, and the rapid mutation rate of PRRSV coupled with the global prevalence of the disease has made it challenging to identify new and emerging variants. Thus, within the lineage system, a dynamic fine-scale classification scheme is needed to provide better resolution on the relatedness of PRRSV-2 viruses to inform disease management and monitoring efforts and facilitate research and communication surrounding circulating PRRSV viruses. Here, we compare potential fine-scale systems for classifying PRRSV-2 variants (i.e., genetic clusters of closely related ORF5 sequences at finer scales than sub-lineage) using a database of 28,730 sequences from 2010 to 2021, representing >55% of the U.S. pig population. In total, we compared 140 approaches that differed in their tree-building method, criteria, and thresholds for defining variants within phylogenetic trees using TreeCluster. Three approaches produced epidemiologically meaningful variants (i.e., ≥5 sequences per cluster), and resulted in reproducible and robust outputs even when the input data or input phylogenies were changed. In the three best performing approaches, the average genetic distance amongst sequences belonging to the same variant was 2.1-2.5%, and the genetic divergence between variants was 2.5-2.7%. Machine learning classification algorithms were also trained to assign new sequences to an existing variant with >95% accuracy, which shows that newly generated sequences could be assigned without repeating the phylogenetic and clustering analyses. Finally, we identified 73 sequence-clusters (dated <1 year apart with close phylogenetic relatedness) associated with circulation events on single farms. The percent of farm sequence-clusters with an ID change was 6.5-8.7% for our best approaches. In contrast, ~43% of farm sequence-clusters had variation in their RFLP-type, further demonstrating how our proposed fine-scale classification system addresses shortcomings of RFLP-typing. Through identifying robust and reproducible classification approaches for PRRSV-2, this work lays the foundation for a fine-scale system that would more reliably group related field viruses and provide better improved clarity for decision-making surrounding disease management.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Phylogenetic-based methods for fine-scale classification of PRRSV-2 ORF5 sequences: a comparison of their robustness and reproducibility

VanderWaal,

Pamornchainvakul,

Kikuti

et al. 2024

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…These features were VP1 amino acid distance and site-wise differences in amino acid positions (4, 13, 24, 32, 33, 43, 48, 49, 57, 69, 97, 100, 124, 135, 139-141, 143, 145, 149-151, 154, 156, 159, 166, 173, 175, 195, 198, 199, 210, 213, 214) Additionally, from the random forest model, model features were ranked in their importance to the performance of our model using mean decrease in model accuracy when a features' data were randomized relative to the outcome (the relative prediction strength of a variable) and improvement in Gini index (measure of node impurity associated with a variable) when data were split on a variable (37,38). This allowed us to highlight highly ranked amino acids in our VP1data, and considered the relative importance of different amino acid sites based on their role in improving model accuracy and node purity in outcome classification (15,39).…”

Section: Training and Testing Machine Learning Modelsmentioning

confidence: 99%

Machine learning approaches for estimating cross-neutralization potential among FMD serotype O viruses

Makau,

Arzt,

VanderWaal

2024

Preprint

Self Cite

View full text Add to dashboard Cite

In this study, we aimed to develop an algorithm that uses sequence data to estimate cross-neutralization between serotype O foot-and-mouth disease viruses (FMDV) based on r1 values, while identifying key genomic sites associated with high or low r1 values. The ability to estimate cross-neutralization potential among co-circulating FMDVs in silico is significant for vaccine developers, animal health agencies making herd immunization decisions, and disease preparedness. Using published data on virus neutralization titer (VNT) assays and associated VP1 sequences from GenBank, we applied machine learning algorithms (BORUTA and random forest) to predict potential cross-reaction between serum/vaccine-virus pairs for 73 distinct serotype O FMDV strains. Model optimization involved tenfold cross-validation and sub-sampling to address data imbalance and improve performance. Model predictors included amino acid distances, site-wise amino acid polymorphisms, and differences in potential N-glycosylation sites. The dataset comprised 108 observations (serum-virus pairs) from 73 distinct viruses with r1 values. Observations were dichotomized using a 0.3 threshold, yielding putative non-cross-neutralizing (< 0.3 r1 values) and cross-neutralizing groups (≥ 0.3 r1 values). The best model had a training accuracy, sensitivity, and specificity of 0.96 (95% CI: 0.88-0.99), 0.93, and 0.96, respectively, and an accuracy of 0.94 (95% CI: 0.71-1.00), sensitivity of 1.00, and specificity of 0.93, positive, and negative predictive values of 0.60 and 1.00, respectively, on one testing dataset and an accuracy, AUC, sensitivity, specificity, and predictive values all approaching 1.00 on a second testing dataset. Additionally, amino acid positions 48, 100, 135, 150, and 151 in the VP1 region alongside amino acid distance were found to be important predictors of cross-neutralization. Our study highlights the value of genetic/genomic data for informing immunization strategies in disease management and understanding potential immune-mediated competition amongst related endemic strains of serotype O FMDVs in the field. We also showcase leveraging routinely generated sequence data and applying a parsimonious machine learning model to expedite decision-making in selection of vaccine candidates and application of vaccines for controlling FMD, particularly serotype O. A similar approach can be applied to other serotypes.

show abstract

Linear epitopes of PRRSV-1 envelope proteins ectodomains are not correlated with broad neutralization

Castillo-Pérez,

Martínez-Lobo,

Frómeta

et al. 2024

Porc Health Manag

View full text Add to dashboard Cite

Background Neutralizing antibodies against PRRSV are capable of conferring protection against viral reinfection, but they tend to be strain specific and usually have poor cross-reactivity. Nonetheless, it has been described that there are individuals capable of efficiently neutralizing viruses of different origin, so it is expected that there are conserved neutralizing epitopes relevant for broad neutralization. However, although immunodominant regions and neutralizing epitopes have been described in different envelope proteins, their role in broad neutralization is unknown. The main objective of this study was to determine whether the linear epitopes existing in the ectodomains of PRRSV envelope proteins play a role in cross-neutralization. Results A pepscan analysis was carried out using synthetic peptides against the ectodomains of PRRSV envelope proteins and PRRSV-hyperimmune sera of different cross-reactivity. The results obtained confirm the existence of antigenic regions in the ectodomains of the GP2, GP3, GP4 and GP5 that tend to be relatively conserved among different PRRSV isolates. Nonetheless, these antigenic regions have poor immunogenicity since they are only recognized by a limited number of sera. Furthermore, no differences were found between the reactivity of sera with broad cross-neutralization capacity and sera with poor heterologous neutralization activity, which indicate that linear epitopes existing in the ectodomains of PRRSV envelope proteins are not relevant for the development of broadly reactive neutralizing antibodies. Subsequently, some selected peptides were used in competition assays with the virus for binding to the cell receptors and in seroneutralization inhibition assays by incubation with hyperimmune sera. Firstly, some peptides that interfere with virus infectivity were identified in competition assays, but only in the case of one viral isolate, which points to the possible existence of a strain-dependent inhibition. However, the results of the seroneutralization inhibition assay indicate that, under the conditions of our study, none of the peptides used was capable of inhibiting virus neutralization by the hyperimmune sera. Conclusions The results obtained indicate that the linear peptides analyzed in this study do not play a major role in the induction of broadly reactive neutralizing antibodies, which could probably depend on conformational neutralizing.

show abstract

Predicting Antigenic Distance from Genetic Data for PRRSV-Type 1: Applications of Machine Learning

Cited by 4 publications

References 89 publications

Phylogenetic-based methods for fine-scale classification of PRRSV-2 ORF5 sequences: a comparison of their robustness and reproducibility

Phylogenetic-based methods for fine-scale classification of PRRSV-2 ORF5 sequences: a comparison of their robustness and reproducibility

Machine learning approaches for estimating cross-neutralization potential among FMD serotype O viruses

Linear epitopes of PRRSV-1 envelope proteins ectodomains are not correlated with broad neutralization

Contact Info

Product

Resources

About