"This is... strange."
Lina murmured, staring at the screen.
"What's wrong?" Eiji approached.
"Docking scores and measured activities have no correlation at all."
Akira checked the data. "The compound with the highest score is actually inactive?"
"Yes. Meanwhile, this compound with a low score has nM-order activity."
Eiji laughed. "The scoring function is crying."
"This isn't funny," Lina looked troubled.
Akira began analyzing. "Which scoring function did you use?"
"AutoDock's Vina. A standard choice, but..."
"Force field-based. Centered on enthalpy terms, entropy simplified."
Eiji added, "Water contribution is also an implicit solvent model, right?"
"Yes. Explicit water molecules aren't considered."
Lina tried another scoring function. "Then, with a knowledge-based score..."
After a few minutes, results appeared.
"Slightly better, but still not perfect."
Akira said, "Each scoring function has strengths and weaknesses."
"For example?"
"Force field-based is physically accurate but computationally expensive. Knowledge-based is fast but dependent on training data."
Eiji organized. "What about empirical scores?"
"Create regression equations from experimental data. Flexible but weak at extrapolation."
Lina proposed a new idea. "Then how about consensus scoring?"
"A method to integrate results from multiple scoring functions," Akira nodded. "Worth trying."
Eiji suggested, "But first, we should understand why the score was wrong."
Lina opened a failure case. "This compound has the highest score but is inactive."
Akira analyzed the structure. "Ah, this is... an entropy penalty."
"What do you mean?"
"This compound is too flexible. It loses significant conformational entropy upon binding. But Vina doesn't evaluate that sufficiently."
Eiji supplemented. "Molecules with high degrees of freedom are enthalpically favorable but entropically unfavorable."
Lina understood. "So the case where the scoring function laughed was underestimating entropy contribution?"
"Yes. Meanwhile, the crying case, or underestimation..."
Akira looked at another compound. "This one coordinates with metal."
"Ah, right. It forms coordination bonds with zinc ions."
"Vina can't accurately handle metal coordination. That's why the score was low."
Eiji said, "Knowing the tool's limits is important."
Lina wrote in her notebook. "Scoring function's smile: ignoring entropy, simple interactions."
"Crying face: metal coordination, explicit water effects, long-range electrostatic interactions."
Akira added, "And handling protonation states is also difficult."
"pH-dependent binding?"
"Yes. If you don't accurately predict charge states at physiological pH, scores go wrong."
Eiji made a practical proposal. "So what do we do? Not trust scores?"
"Trust, but don't blindly believe," Lina answered. "Scores are one indicator. Combine with other data."
Akira agreed. "Experimental data, physicochemical properties, synthetic accessibility. Judge comprehensively."
Lina started a new analysis. "I'll try interaction fingerprint methods."
"Not score values, but looking at interaction patterns," Eiji understood.
"Right. Qualitatively evaluate patterns of hydrogen bonds, hydrophobic contacts, electrostatic interactions."
Akira looked at the screen. "Similarity to known active compounds?"
"High. Interacting with the same amino acid residues."
Eiji laughed. "Even when the scoring function cries, the interaction pattern smiles."
Lina organized. "So relative patterns are more important than absolute score values?"
"Case by case," Akira said. "But multifaceted evaluation is always important."
Eiji concluded. "Perfect scoring functions don't exist. That's why human judgment is necessary."
Lina smiled. "Whether scoring functions laugh or cry, we analyze calmly."
"That's the reality of computational drug design," Akira said.
The three continued discussing methods to make sound judgments without drowning in the sea of scores.