"I want to express this protein in much larger quantities."
Eiji said while looking at culture data.
"Then let's optimize the promoter," Mikhail suggested.
"Promoter?" Sena asked.
"The switch for gene expression. Determines the transcription start point."
"How do you optimize it?"
Mikhail began explaining. "Promoters consist of multiple elements."
A DNA sequence appeared on screen.
"First, the TATA box. A sequence recognized by RNA polymerase."
"TATAAA..." Sena read.
"Yes. That's the core element. But this alone gives low expression."
"What else is needed?"
"Enhancers or silencers. Sequences where transcription factors bind."
Eiji supplemented. "Transcription factors are proteins that regulate promoter activity."
"There are activating factors and repressing factors."
Sena wrote in her notebook. "So should we increase binding sites for activating factors?"
"Theoretically yes, but it's not that simple," Mikhail said.
"Why?"
"Interactions between transcription factors, DNA secondary structure, nucleosome placement... everything influences it."
"Too complex."
"So we predict with machine learning."
Mikhail showed a model. A neural network predicting expression level from sequence.
"Using this model to design high-expression promoters."
"How?"
"Solve as an inverse problem. Set target expression level and search for sequences achieving it."
Eiji showed interest. "Optimization algorithm?"
"Genetic algorithm. Mutate and evolve sequences."
"Interesting."
Mikhail demonstrated. Input initial sequence. Set target expression to 10-fold.
"Start optimization."
The sequence changed on screen. Predicted expression values increased each generation.
"After 100 generations..."
A new sequence displayed.
"Predicted expression is 12-fold. Target achieved."
Sena was impressed. "Amazing. But will this sequence actually increase expression?"
"It's a prediction, so not certain. Need experimental confirmation."
Eiji presented another perspective. "Should we also consider tissue specificity?"
"Right. The optimal promoter differs depending on whether you want high expression in liver or brain."
"Because of tissue-specific transcription factors."
Sena asked. "So you train models for each tissue?"
"Ideally yes. But some tissues have scarce data."
"In that case?"
"Transfer learning. Train on tissues with abundant data, then fine-tune on tissues with little data."
"I see."
Mikhail warned. "But promoter design also has constraints."
"Like what?"
"Avoiding CpG islands. Sequences prone to methylation have suppressed expression."
"Methylation?"
Eiji explained. "Chemical modification of DNA. One type of epigenetic control."
"Difficult..."
"So it becomes multi-objective optimization. Maximize expression while suppressing CpG content."
Sena summarized. "Promoter design requires considering multiple factors simultaneously."
"Yes. That's why it's an adventure," Mikhail smiled.
"Adventure?"
"Exploring unknown sequence space. Don't know what you'll find."
Eiji continued. "But with data and models, much more efficient than blind search."
"The era of synthetic biology."
Sena said thoughtfully, "Being able to design gene switches ourselves."
"But responsibility comes with it," Mikhail became serious.
"Responsibility?"
"Designed sequences might have unexpected effects. Like off-target expression."
"So design carefully and validate thoroughly."
Eiji nodded. "The more powerful the technology, the more carefully we must handle it."
Mikhail concluded. "DNA promoter optimization is the fusion of science and engineering."
"And an adventure," Sena murmured.
A small sequence called a promoter. But it greatly influences life's expression. Its design is truly an adventure.