"Why won't this ligand bind?"
Sena showed a confused expression.
"The docking score is perfect."
Lina looked at the screen. "The Induced-Fit trap."
"Induced-Fit?"
"Proteins change shape when ligands approach. But normal docking uses rigid structures."
Sena tried to understand. "So it's different from the actual binding shape?"
"Yes. Docking with the pre-binding structure won't find the optimal post-binding arrangement."
Mikhail joined in. "The lock-and-key metaphor is insufficient here."
"What do you mean?"
"The lock is flexible and deforms when the key arrives. They adapt to each other."
Lina showed an example. "This is the apo structure. Without ligand bound."
"And this is the holo structure. With ligand bound."
Overlaying the two structures revealed clear differences.
"The loop... moved 15 angstroms," Sena was surprised.
"Yes. Such large changes can't be handled by rigid docking."
Mikhail organized the problem. "The Induced-Fit trap has two forms."
"The first?"
"The binding pocket is closed. The ligand physically can't enter."
"The second?"
"The pocket is open but the shape isn't optimal. Scores come out low."
Lina added, "Both create false negatives. Actually binds, but prediction misses it."
Sena asked, "Then how do we solve it?"
"Several methods," Lina began explaining. "First, flexible docking."
"Dock while moving protein side chains?"
"Yes. But computational cost increases exponentially."
Mikhail supplemented. "Moving all residues isn't realistic. Select only important residues."
"Which residues are important?"
"Residues in the binding pocket, especially hydrophilic ones. Side chain conformations change easily."
Lina introduced another method. "Next, ensemble docking."
"Ensemble?"
"Dock against multiple protein structures. Like structures generated from MD simulation."
Sena understood. "Try various shapes and find the best one."
"Correct. But this is also computationally intensive."
Mikhail made a practical proposal. "So a staged approach is better."
"Staged?"
"First, narrow candidates with rigid docking. Next, flexible docking only for promising candidates."
Lina agreed. "Balance of efficiency and accuracy."
Sena had another question. "But which structure should we use? Apo? Holo?"
"Difficult question," Lina admitted. "For novel ligands, there's no holo structure."
"Then can we only use apo structure?"
"That's where homology modeling helps," Mikhail said.
"Predict the target's holo structure by referencing similar proteins' holo structures."
Lina showed a real example. "This target has no holo structure, but there's a 60 percent homologous protein's holo structure."
"Use that as template to predict structure."
Sena was impressed. "But the accuracy?"
"Not perfect. But often better than apo structure."
Mikhail introduced another strategy. "Can also use AlphaFold."
"Structure prediction?"
"Yes. Can directly predict ligand complex structures too."
Lina cautioned. "But AlphaFold depends on training data. Weak at novel binding modes."
"So combine multiple methods," Mikhail concluded.
Sena organized in her notebook. "Methods to avoid Induced-Fit trap: flexible docking, ensemble, homology modeling, AlphaFold."
"And," Lina added, "experimental validation is ultimately necessary."
"X-ray crystallography?"
"That or Cryo-EM. Determine actual complex structure."
Sena said, "Computation is prediction, experiment is truth."
"Good expression," Lina smiled. "But without computation, experimental direction isn't clear."
Mikhail said philosophically, "Induced-Fit demonstrates life's flexibility."
"Not fixed structures, but dynamic adaptation."
"Yes. Understanding and modeling that complexity is our challenge."
Sena was newly impressed by the dynamic beauty of proteins. Induced-Fit is a trap, but also the essence of life.