"Until now, we were the targets. The side where drugs bind."
Lina said quietly.
"But now, we ourselves are being designed," Mikhail showed the screen.
Akira was surprised. "Designing proteins from scratch?"
"Yes. Specify a structure, and AI generates sequences to realize it."
Lina tried to understand. "The inverse folding problem?"
"Correct. Normally, we predict structure from sequence. But this designs sequence from structure."
Akira opened his notebook. "What's the mechanism?"
Mikhail drew a diagram. "Deep generative models. Encode structure, decode sequence."
"VAE? GAN?"
"Recently, diffusion models and Transformers are mainstream."
Lina asked, "Training data?"
"Protein Data Bank. Hundreds of thousands of protein structure-sequence pairs."
"With that much data, patterns can be learned," Akira was convinced.
Mikhail showed a real example. "This is a completely new protein. Doesn't exist in nature."
"But it folds stably?"
"Confirmed experimentally. Took the predicted structure."
Lina was moved. "It truly understands the language of life."
"Not perfect yet," Mikhail said humbly. "But progressing rapidly."
Akira confirmed practicality. "What can it be used for?"
"Enzyme design, antibody design, binding protein design."
"Concrete examples?"
Mikhail opened another screen. "This is a miniprotein that binds COVID spike protein."
"Small... only 60 residues?"
"Yes. But binds with nanomolar affinity. Computationally designed, experimentally validated."
Lina was excited. "Much smaller than antibodies, but specific."
"Manufacturing is also easy. Can express in E. coli."
Akira offered another perspective. "But how do you design function? Not just structure."
"Good question," Mikhail nodded. "That's the next challenge."
"The relationship between structure and function?"
"Complex. Same structure, but slight sequence differences change function."
Lina proposed, "To design catalytic activity?"
"Specify active site geometry. Place specific residues there."
Mikhail showed a design example. "This enzyme hydrolyzes esters."
"How did you design it?"
"First, determine catalytic triad to stabilize transition state. Next, design scaffold to realize that arrangement."
Akira understood. "Build structure with function as priority."
"Yes. Software like Rosetta makes this possible."
Lina asked about another possibility. "Therapeutic proteins?"
"Of course. For example, binders that bind specific targets."
"Instead of antibodies?"
"Smaller, more stable, easier to manufacture than antibodies. A major application of de novo design."
Akira pointed out a practical problem. "But immunogenicity?"
"A challenge. Novel sequences risk being recognized by the immune system."
Mikhail explained. "So we design considering humanization. Choose sequences close to natural proteins."
Lina asked another question. "Can membrane proteins be designed too?"
"In principle, yes. But still difficult due to membrane environment complexity."
"What's difficult?"
"Interaction with lipid bilayer, transmembrane helix placement, intracellular transport."
Akira said, "More constraints than soluble proteins."
"Yes. But recently, deep learning models specialized for membrane proteins have emerged."
Lina imagined the future. "Someday, can we design any protein?"
"Possible," Mikhail said quietly. "But humility is also needed."
"Why?"
"Nature has optimized proteins over billions of years. We've only been at it for a few years."
Akira agreed. "Learn from nature, surpass nature. That's the goal."
Mikhail showed a new project. "This is the next challenge. Designing multifunctional proteins."
"One protein with multiple functions?"
"Yes. Combining diagnosis and therapy, or acting on multiple targets simultaneously."
Lina laughed. "When proteins become 'the designed,' possibilities explode."
"But," Mikhail's face became serious, "responsibility is also great. Because we're designing life."
Akira nodded. "Ethical discussions are also necessary."
"Yes. Not just technology, but wisdom too."
The three gazed at the new horizon of protein design, cautiously but hopefully.