"Same protein, but different mRNA changes expression by a factor of 100."
Mikhail said quietly.
"A hundred times?" Sena was surprised. "Even though the amino acid sequence is the same?"
"Yes. The choice of codons dramatically changes translation efficiency."
Lina showed interest. "Codon optimization."
"Precisely, it's more than that. Not just codons. RNA secondary structure, 5'UTR, 3'UTR... everything affects translation efficiency."
Sena opened her notebook. "Please teach me in order."
"First, codons. Three-base sequences encoding amino acids."
"I know that."
"But codons are degenerate. Leucine has six codons."
"Six... can you choose any?"
"Theoretically yes. But translation efficiency changes."
Mikhail showed a table. Usage frequency of each codon.
"Each organism has preferred codons. Codon usage bias."
"If you want high expression in humans, use human-preferred codons."
Lina supplemented. "That corresponds to tRNA abundance."
"tRNA?"
"Transfer RNA. Recognizes codons and carries amino acids."
"Using codons corresponding to abundant tRNAs makes translation fast."
Sena understood. "So even with the same amino acid sequence, expression levels change."
"Yes. But that's not all," Mikhail continued.
"What else?"
"RNA secondary structure. mRNA is single-stranded but folds on itself."
"Like hairpin structures?"
"Yes. Especially if there's strong secondary structure near the 5' end, ribosomes can't bind."
"Translation doesn't start..."
"So we design to avoid secondary structure in the 5'UTR and near the translation start codon."
Lina asked. "How do you predict secondary structure?"
"Computational tools. Like RNAfold. Predicting structure by free energy minimization."
"So you choose codons to avoid low-energy structures?"
"Correct. Calculate repeatedly to find the optimal sequence."
Sena had another question. "But changing codons also changes secondary structure, right?"
"Yes. So it becomes a multi-objective optimization problem."
"Multi-objective?"
Mikhail explained. "Maximize codon usage frequency while minimizing secondary structure. Find sequences satisfying both."
"Sounds difficult..."
"That's why machine learning comes in."
Lina's eyes sparkled. "Designing mRNA with AI?"
"Yes. Training models from known mRNA sequences and expression level data."
"What kind of model?"
"Like convolutional neural networks. Learning sequence patterns."
Sena was impressed. "AI proposes optimal mRNA sequences?"
"Ideally yes. But not perfect yet."
"What's difficult?"
"Too many factors affecting expression. Codons, secondary structure, UTRs, plus mRNA stability, subcellular localization..."
Lina supplemented. "And it varies by cell type."
"Yes. The optimal sequence for liver cells isn't optimal for neurons."
Sena sighed. "Too complex."
"But with recent mRNA vaccine success, this field is rapidly advancing," Mikhail said.
"COVID-19 vaccines?"
"Yes. Those are combinations of codon optimization and modified nucleotides."
Lina explained in detail. "Using pseudouridine, a modified base, to suppress immune response while increasing translation efficiency."
"Amazing technology."
Mikhail smiled. "The era of mRNA therapeutics is coming. Instead of administering protein directly, we administer its blueprint."
"Designing the blueprint," Sena murmured.
"Yes. The translation efficiency blueprint hidden in mRNA sequences. Reading and rewriting it."
Lina said, "Feels like editing life's program."
Mikhail nodded. "That's exactly why we must design carefully."
Translation efficiency, an invisible property. But controllable in the form of sequence. That's the appeal of mRNA design.