Multi-Epitope mRNA Vaccine Design | Science Fair Ideas

ISEF Category: Biomedical and Health Sciences

Subcategory: Immunology  ·  Difficulty: Advanced  ·  Setup: Home Setup  ·  Time: 1 to 2 Months

The Hook

Candida auris is a yeast that ignores most of the antifungals doctors have. It spreads in hospitals, kills about one in three people who get a bloodstream infection, and there is no licensed vaccine. You will not cure it in a science fair. You can, however, design the messenger RNA blueprint for a vaccine candidate, score it with public tools, and explain on paper why your design is worth a wet-lab follow-up.

What Is It?

An mRNA vaccine teaches your immune system to recognize a pathogen by handing it a small genetic recipe. Your ribosomes read the mRNA, make the matching piece of protein, and show that piece to immune cells. A multi-epitope construct skips the full protein and stitches together only the parts most likely to be recognized by T cells and B cells across many people.

Designing one in silico is a stack of small decisions. First, pick antigens from Candida auris that the immune system can actually see. Second, predict which short fragments (epitopes) bind common human HLA types so the vaccine works across populations. Third, link those epitopes with short spacer sequences, optimize the codons so a human ribosome translates the mRNA efficiently, and check that the resulting RNA folds into a stable shape using RNAfold.

The whole project is computational. You work in a notebook, store your designs in plain text files, and produce a final ranked construct with a clean justification for every choice.

Why This Is a Good Topic

Fungal vaccines are an open problem in medicine, so a careful in-silico design is genuinely useful prior art. The pipeline is fully measurable: binding scores, population coverage percentages, GC content, free energy of folding. A judge can question any step and you can show the number behind the choice. A high school student can realistically learn epitope prediction, codon usage, and RNA structure scoring in a couple of months without any lab access.

Research Questions

• How does the predicted population coverage of your construct change when you require binding to more HLA alleles?
• What is the effect of changing the linker sequence between epitopes on the predicted folding stability of the final mRNA?
• Does codon optimization for human expression always improve the predicted secondary-structure stability, or does it sometimes hurt it?
• To what extent do epitopes predicted from a surface antigen overlap with epitopes predicted from a secreted antigen of Candida auris?
• Which spacer length between epitopes gives the best balance of folding stability and predicted antigen processing?

Basic Materials

• Laptop with at least 8 GB of RAM and 30 GB of free disk space.
• Stable internet connection for using web-based prediction servers.
• Digital lab notebook to record every input file, parameter, and score.
• Free GitHub account to version your construct files and analysis scripts.

Advanced Materials

• Workstation or cloud VM for running local copies of NetMHCpan, NetMHCIIpan, or IEDB tools.
• Local install of ViennaRNA for batched RNAfold runs on many candidate constructs.
• Population HLA frequency tables from the Allele Frequency Net Database.
• Curated Candida auris proteomes downloaded from UniProt and FungiDB.

Software & Tools

• IEDB Analysis Resource: Free web tools for predicting MHC class I and class II epitopes and estimating population coverage.
• ViennaRNA (RNAfold): Computes minimum free energy and predicted secondary structure of your final mRNA construct.
• Biopython: Parses protein and nucleotide sequences, builds constructs, and computes codon usage statistics.
• Codon Usage Database: Reference tables of human codon frequencies used as the target for codon optimization.
• Python: Glues the pipeline together and produces ranking tables and plots for your final report.

Experiment Steps

1. Pick two or three Candida auris proteins that are surface-exposed or secreted and justify each choice from the literature.
2. Decide your HLA panel and your binding threshold before you run any prediction so the scoring is fair across antigens.
3. Build a candidate construct by stitching predicted epitopes with one or more linker designs you defined in advance.
4. Plan a codon optimization rule that targets human expression while avoiding accidental hairpins in the mRNA.
5. Run RNAfold on every candidate construct and rank them on a clear, written-down scoring rule.
6. Compare your top construct against a single-antigen baseline and document every parameter you tuned and why.

Common Pitfalls

• Choosing antigens because they appear in many papers, not because they are predicted to be accessible to the immune system, which biases the whole pipeline.
• Using a single HLA allele for epitope prediction, which makes your construct look strong on paper but useless across real populations.
• Forgetting to mask predicted epitopes that closely match human self-peptides, which is the in-silico version of an autoimmunity risk.
• Optimizing codons for the highest possible expression and ignoring the GC content spike that wrecks RNAfold stability.
• Comparing construct designs using different scoring thresholds, which hides the fact that one design only wins because the bar was lower.

What Makes This Competitive

A class-level version stitches a few epitopes together and reports a folding score. A competitive ISEF-level version commits to a written scoring rubric up front, sweeps construct designs across multiple linker lengths and codon strategies, and reports population coverage across globally representative HLA frequencies. Add a self-similarity filter against the human proteome to flag autoimmunity risk and a sensitivity analysis that varies one parameter at a time. Close with a head-to-head comparison against a published single-antigen baseline.

Project Variations

• Repeat the design pipeline for a different drug-resistant fungal pathogen, such as Aspergillus fumigatus, and compare which step in the pipeline drives most of the score difference.
• Swap the multi-epitope construct for a full-length antigen design and ask whether the population coverage really drops as much as the literature suggests.
• Add a stability stress test that mutates one or two nucleotides in your top construct and reports how often the folding score collapses.

Learn More

• IEDB Analysis Resource: Free web-based epitope prediction and population coverage tools with full documentation.
• UniProt: Free protein sequence database where you can pull annotated Candida auris proteins for your antigen pool.
• FungiDB: Free fungal genomics portal with curated Candida auris genomes and gene expression data.
• ViennaRNA Web Services: Free RNAfold tool plus tutorials on interpreting minimum free energy and base-pair probabilities.
• PubMed: Search review articles on multi-epitope vaccine design and fungal immunology for background and citations.
• MIT OpenCourseWare: Free introductory immunology and computational biology lectures that cover the supporting concepts.