Preparing for a Future Pandemic: Analyzing Emerging Sequence Variations in Avian Influenza

Thank you for watching my video!

Transcript:

Hello, my name is Sehar Sidhu, and I'm a grade 10 student fighting against the spread of dangerous viruses.

The intercontinental spread of  Avian Influenza H5N1 has caused unprecedented mortality in mammals and infection in poultry, dairy cows and humans, including a 13-year-old teenager in BC, Canada.

Given the reliance on retrospective mutation analysis, we need to be more ready for the upcoming pandemic without delays in detecting new variants.

With the help of various experts, I formulated a computational platform to analyze sequence variations and entropy in HA, NA and PB2, important viral proteins in the circulating genotypes B3.13, and D1.1, with bovine and avian origin respectively.

I was able to predict mutational hotspots with their role in tropism, viral fitness, replication and drug resistance.

In conclusion, this machine learning based model will be able to rapidly detect mutations by combining retrospective and predictive analysis. This will allow for faster containment and the rapid design of vaccines against potential pandemic strains with higher rates of transmission and serious illness.

Introduction:

The world is witnessing a panzootic of highly pathogenic avian influenza (HPAI) H5N1 characterized by (Figures 1-4):

(1)   Rapid spread across all continents other than Australia,

(2)   Frequent genomic reassortment,

(3)  Spillover infections expanding the host range to wild mammals (raccoons, weasels, rabbits) and domestic mammals (dairy cattle, cats, dogs), increasing the risk of spillover infections in humans (Peacock et al., 2025).

Current circulating B3.13 and D1.1 genotypes of H5N1 2.3.4.4b are known to infect humans (WHO).

This raises an important question:

How can computational methods help to detect and predict the emergence of novel mutations that have the potential to increase the risk of spillover infections and the efficient transmission of novel variants capable of causing serious disease in humans?

Problem:

Current reliance on retrospective analysis of sequence data introduces an undesirable delay in detection of potentially dangerous viral variants capable of causing higher rates of transmission & serious illness.

Solution:

Computational methods can help to analyze the patterns of changes in viral genome sequences over time to detect the emergence of clusters of mutations of novel variants with the potential for increasing transmissibility and pathogenesis.

Goal:

To develop a computational framework to analyze the sequence variations and entropy changes in H5N1 hemagglutinin (HA) and neuraminidase (NA) genes over time.

To predict emerging mutations in H5N1 using statistical mechanics and machine learning to anticipate emergence of viral variants with the potential to spread efficiently and to cause widespread serious illness in a future pandemic.

Methodology:

1. Collected H5N1 protein sequences from H5N1 B3.13 and D1.1 genotypes (clade 2.3.4.4b), from the Global Initiative on Sharing Avian Influenza Data (GISAID), for mammals & avians between 2024-2026.

2.  Pre-processed H5N1 sequences to clean files (e.g., removing duplicate/incomplete entries) and multiple sequence alignment was carried out using the MAFFT_V.11 program.

3. Selected reference sequences: A3 genotype & B3.6 genotype for D1.1 and B3.13 lineages respectively based on phylogenetic relevance and temporal proximity to genotype emergence.

4. Calculated mutation prevalence as the percentage of sequences harboring a specific amino acid substitution within each dataset using R-studio. Aligned amino acid sequences of H5N1 hemagglutinin (HA), neuraminidase (NA) & PB2 proteins were analyzed to assess temporal mutation dynamics across defined time windows.

5. Constructed heatmaps using the Complex Heatmap package in RStudio, with prevalence values represented using a continuous blue color scale (0–100%) and with entropy values using yellow-dark red color scale (0.0-1.0).

6.  Mutational entropy (measures variability at specific site) calculations were carried out using R-studio. The output was displayed as a graph of Entropy vs. Residue Number, representing the mutational entropy at each residue position within the viral protein sequence.

7. Mutational response function (MRF) was calculated using Evolutionary Variation Observation Learning and Variant Exploration or EVOLVE platform (Satyam et al., 2025) to capture mutational state transitions in the evolutionary process as a result of accumulated mutations & as virus adapt in the real time (Figure).

8. Predicted emerging mutations by combining statistical mechanisms with machine learning.

9. Protein Structural Modeling allowed visualization of mutational hotspots on H5N1 structured retrieved from RCSB-Protein Databank (PDB). Identified mutated residues were structurally mapped on D1.1 and B3.13 HA and NA protein and visualized using PyMOL, a molecular visualization system.

RESULTS:

Sequence analysis, entropy plots, structural modelling, mutational response function (MRF) and mutation prediction of current panzootic caused by H5N1 clade 2.3.4.4b were examined between 2024-2026 in B3.13 and D1.1 genotypes with mutation prevalence (%) as shown below (Figures 1-3):

Legend:

M-Mammalian Host

A- Avian Host

N- Number of Sequences

B3.13 HA

Substitutions in avian (N=464) and mammals (N=3641):

-   D104G (70%; A,M) and V147M (30% A; 39%M) substitutions enabled the B3.13 genotype to engage avian NeuAc (α2,3) and mammalian NeuGc (α2,3) receptors, expanding receptor tropism, improving viral replication in the bovine mammary gland & promoting viral fitness in dairy cattle (Hassard et al., 2026).

- S336N (71% A, 68% M) is reported as a virulence marker, with an increased prevalence in viruses isolated from dairy cows (Nguyen et al., 2025).

D1.1 HA

Substitutions in avian (N=6062) and mammals (N=213):

-    High mutation prevalence (93%) in mammals and high entropy from Jul 2025-Mar 2026 for T52A at antigenic site-E suggests ongoing selection for immune-evasive mutations.

B3.13 NA

Substitutions in avian (N=468) and mammals (N=3666):

-     High prevalence (99.9%) substitution such as V67I suggests likely becoming “fixed” in mammals.

D1.1 NA

Substitutions in avian (N=6064) and mammals (N=214):

-         N247S substitution in mammals and birds displayed ~99% high mutational prevalence and low entropy (2024-2026). Mutation at 247 position could reduce affinity for oseltamivir, an anti-viral drug (Pokorná et al., 2018).

B3.13 PB2

Substitutions in avian (N=465) and mammals (N=3577):

-    Importantly, M631L with ~99% mutation prevalence in B3.13-PB2 avian and mammals acts as a critical driver of increased polymerase/replication activity with expanding host range and cross-species transmission (e.g., avian-to-cattle) of the H5N1 virus (Zhang et al., 2026).

D1.1 PB2

Substitutions in avian (N=5999) and mammals (N=203):

- D1.1 PB2 in mammals exhibited low mutational prevalence for E627K (23.2%) and D701N (22.2%) reported for increased polymerase/replication activity, increasing viral fitness, host adaptability, pathogenicity and enhanced transmission in mammals (Hu et al., 2024).

As human adaptation marker, D1.1 PB2 E267K was detected in a Canadian 13-year-old girl from British Columbia (Jassem et al., 2025).

Mutation-Response-Function (MRF) and mutation prediction of B3.13 and D1.1 HA, NA and PB2

For D1.1, MRF plots show the synchronous peak for HA, NA and PB2 at ~0.27 between May and August 2025 and return by Oct-Jan26 constituting a "genomic phase transition" (Figure 4). This indicates increase in exploration for virus entry-exit (HA-NA) and for replication machinery (PB2) in response to spillover & viral transmission events (FAO et al., 2025).

.

For B3.13, coherent Mar25-Jun25 jump in HA, NA, and PB2 followed by the secondary rise through Jun25-Oct25, indicates different evolutionary trajectory of B3.13 from D1.1 in real time.

Finally, figure 5 shows the surface view of emerging mutations highlighted in yellow using the multilayer perceptron machine learning model for B3.13 and D1.1.

Significance on animal and public health:

Since the outbreak in US dairy cows, March 2024, 71 human cases of HPAI (H5N1) have been reported in the Canada and United States caused by either the B3.13 genotype circulating in dairy cows or the D1.1 genotype circulating in birds (Morse et al., 2024).

These recent fatal & severe cases of infections in humans caused by D1.1 H5N1 raises pandemic concern with human transmission.

Developed machine learning based diagnostic and predictive novel computational framework that will assist with the three major areas of influenza research (Figures):

(1) Genomic signatures of pathogenicity and transmission

By rapidly identifying H5N1 clade2.3.4.4b genotype and host specific set of adaptive mutations in HA, NA, and PB2 that play central roles in the virulence and pathogenesis of H5N1.

(2) Surveillance

Notably, framework provides early detection of hotspot mutations, & emergence of mutational state transitions before widespread transmission occurs.

(3) Therapies and pandemic preparedness

Emergence of resistance mutations at antigenic sites on binding sites of antiviral suggests development of multi-target antiviral drugs as control strategies for a "moderate risk" potential future pandemic emergence (CDC).

In conclusion, this machine learning based model will be able to rapidly detect mutations by combining retrospective and predictive analysis. This will allow for faster containment and the rapid design of vaccines against potential pandemic strains with higher rates of transmission and serious illness.

Future Directions:

Multi-gene approach - Expanding sequence analysis & prediction to other gene segments (e.g., matrix proteins).

Multi-database approach - Viral isolates were obtained from GISAID with limitations/biases. Farm animals such as cattle and chicken are most common in B3.13 and D1.1 respectively.

Model enhancement approach - Improving the model by training with wider subtypes e.g. H1N1, H7N9 etc.

Functional experimental validation - Testing in lab of identified adaptive or hotspot mutations to determine the resulting virus pathogenicity & transmissibility.

Finally, this will allow moving from "reactive culling" to "predictive prevention" and advocating through the lens of "One Health" approach. (Figure)

Thank you for your interest and time!

I am deeply thankful to Prof. Kenneth Ng & Ana Podadera, postdoctoral-fellow at the Department of Chemistry & Bio-chemistry from the University of Windsor, ON for guiding & working with me on this project.

Prof. Ng with Biochemistry Ph.D., Stanford University is an expert on viruses and their work provides the foundation for designing novel antiviral therapeutics. Ana was very helpful in guiding with the entropy and mutation frequency heatmaps along with the laboratory based experimentations at the University of Windsor.

I am also very grateful to Satyam Sangeet Ph.D. at the Department of Physics, University of Sydney, Australia for their collaboration on this project for the statistical mechanics and machine learning model of mutation prediction.

I would also like to say "thank you" to the Windsor Regional Science, Technology, & Engineering Fair (WRSTEF) for providing me with this amazing opportunity to go to CWSF!

References

Figures were generated with the help of Canva

CDC or Centers for Disease Control and Prevention. (2024, May 22). IRAT virus summaries. U.S. Department of Health and Human Services. Retrieved from: https://www.cdc.gov/pandemic-flu/php/monitoring/irat-virus-summaries.html

DeLano, W. L. (2002). Pymol: An open-source molecular graphics tool. CCP4 Newsl. protein crystallogr, 40(1), 82-92.

Food and Agriculture Organization of the United Nations (FAO), World Health Organization, & World Organisation for Animal Health. (2025, April 17). Updated joint FAO/WHO/WOAH public health assessment of recent influenza A(H5) virus events in animals and people. World Organisation for Animal Health. Retrieved from: https://www.woah.org/app/uploads/2025/04/2025-04-17-fao-woah-who-h5n1-assessment.pdf

Hassard, J. A., Yang, J., Dadonaite, B., Pekar, J. E., Yu, J., Richardson, S. A. S., Pinto, R. M., Valdez, K. R., Lemey, P., Quantrill, J. L., Xue, J., Masonou, T., Case, K. M., Ajeian, J., Woodall, M. N. J., Ross, R. A., Hudson, N., Zhong, K., Cao, H., ... Peacock, T. P. (2026). Bovine H5N1 influenza viruses have adapted to more efficiently use receptors abundant in cattle. bioRxiv. https://doi.org/10.64898/2026.04.02.715584

Hu, X., Saxena, A., Magstadt, D. R., Gauger, P. C., Burrough, E. R., Zhang, J., … Li, G. (2024). Genomic characterization of highly pathogenic avian influenza A H5N1 virus newly emerged in dairy cattle. Emerging Microbes & Infections, 13(1) https://doi.org/10.1080/22221751.2024.2380421

Jain, R., Sharma, H., Pena, L., Jit, S., Rathi, B., De Oliveira, R. N., & Verma, M. (2025). Influenza virus: Genomic insights, evolution, and its clinical presentation. Microbial Pathogenesis, 205, 107671.

Jassem, A. N., Roberts, A., Tyson, J., Zlosnik, J. E., Russell, S. L., Caleta, J. M., ... & Goldfarb, D. M. (2025). Critical illness in an adolescent with influenza A (H5N1) virus infection. New England Journal of Medicine, 392(9), 927-929.

Katoh, K., Misawa, K., Kuma, K. I., & Miyata, T. (2002). MAFFT: a novel method for rapid multiple sequence alignment based on fast Fourier transform. Nucleic acids research, 30(14), 3059-3066.

Krammer, F., Hermann, E., & Rasmussen, A. L. (2025). Highly pathogenic avian influenza H5N1: history, current situation, and outlook. Journal of Virology, 99(4), e02209-24.

Long, J. S., Mistry, B., Haslam, S. M., & Barclay, W. S. (2019). Host and viral determinants of influenza A virus species specificity. Nature Reviews Microbiology, 17(2), 67-81

Morse, J., Coyle, J., Mikesell, L., Stoddard, B., Eckel, S., Weinberg, M., ... & Bagdasarian, N. (2024). Influenza A (H5N1) virus infection in two dairy farm workers in Michigan. New England Journal of Medicine, 391(10), 963-964.

Nguyen, T. Q., Hutter, C. R., Markin, A., Thomas, M., Lantz, K., Killian, M. L., ... & Anderson, T. K. (2025). Emergence and interstate spread of highly pathogenic avian influenza A (H5N1) in dairy cattle in the United States. Science, 388(6745), eadq0900.

Peacock, T. P., Moncla, L., Dudas, G., VanInsberghe, D., Sukhova, K., Lloyd-Smith, J. O., ... & Nelson, M. I. (2025). The global H5N1 influenza panzootic in mammals. Nature, 637(8045), 304-313.

Ratnadass, A., & Deguine, J. P. (2021). Crop protection practices and viral zoonotic risks within a One Health framework. Science of the Total Environment, 774, 145172.

Salomon, R., & Webster, R. G. (2009). The influenza virus enigma. Cell, 136(3), 402-410.

.

Sangeet, S., Sinha, A., Nair, M. B., Mahata, A., Sarkar, R., & Roy, S. (2025). EVOLVE: A Web Platform for AI-Based Protein Mutation Prediction and Evolutionary Phase Exploration. Journal of Chemical Information and Modeling, 65(9), 4293-4310.

Shannon, C. E. (1948). A mathematical theory of communication. The Bell system technical journal, 27(3), 379-423.

Shu, Y., & McCauley, J. (2017). GISAID: Global initiative on sharing all influenza data–from vision to reality. Eurosurveillance, 22(13), 30494.

Signore, A. V., Joseph, T., Ranadheera, C., Erdelyan, C. N., Alkie, T. N., Raj, S., ... & Berhane, Y. (2025). Neuraminidase reassortment and oseltamivir resistance in clade 2.3. 4.4 b A (H5N1) viruses circulating among Canadian poultry, 2024. Emerging Microbes & Infections, 14(1), 2469643. https://doi.org/10.1080/22221751.2025.2469643

WHO. (n.d.). Cumulative number of confirmed human cases for avian (March, 2024). Retrieved from:https://cdn.who.int/media/docs/default-source/influenza/h5n1-human-case-cumulative-table/cumulative-number-of-confirmed-human-cases-for-avian-influenza-a(h5n1)-reported-to-who--2003-2024.pdf

Zhang, G., Shi, Y., Ge, H., Wang, Y., Lu, L., Jiang, S., & Wang, Q. (2025). Genomic signatures and host adaptation of H5N1 clade 2.3. 4.4 b: A call for global surveillance and multi-target antiviral strategies. Current Research in Microbial Sciences, 8, 100377.

Zhang, L., Lai, Y., Cui, Y., Yang, Q., Shao, Y., Ding, S., ... & Deng, T. (2025). Emergence of mammalian-adaptive PB2 mutations enhances polymerase activity and pathogenicity of cattle-derived H5N1 influenza A virus. Nature Communications, 17, 1011.