Kvíz spojený s průvodcem · Střední úroveň
AI in Lead Optimization Quiz
Tests lead optimization cycles, multi-property trade-offs, prediction versus measurement, validation splits, and chemist oversight.
Otázka 1 z 8
What does a model-predicted potency improvement establish?
Náhled každé otázky8 otázky
Odpovědi a vysvětlení se zobrazují, jak budete v kvízu.
- What does a model-predicted potency improvement establish?
- That synthesis will be straightforward
- A hypothesis to verify with synthesized compounds and assays
- A confirmed improvement in patients
- That exposure and toxicity are acceptable
- Why can a more potent analogue be a worse lead?
- A lead is judged only by potency
- A model score measures every clinical property
- It may have poor solubility, exposure, selectivity, or safety
- Potency always worsens solubility
- Which sequence describes a design-make-test cycle?
- Propose structures, synthesize selected compounds, and measure outcomes
- Train a model once and skip experiments
- Select a target and begin clinical use
- Rank compounds only by literature frequency
- Why split test data by scaffold or time?
- To make all compounds chemically identical
- To avoid measuring potency
- To increase predicted scores
- To reduce leakage from close analogues and better assess novelty
- How should a chemist handle an AI-generated structure?
- Use the predicted score as a lab result
- Review its plausibility and test it experimentally
- Assume the model guarantees synthesis
- Advance it because the structure is novel
- Which measurement should be considered alongside potency?
- Solubility, metabolic stability, selectivity, and exposure
- Only the number of atoms
- Only a preferred scaffold name
- The model’s training loss
- How can chemist preference data inform a learning-to-rank model?
- It proves which compound works in patients
- It replaces biological assays
- It automatically predicts every ADME property
- It can train a model to prioritize compounds according to expert comparisons
- Why preserve negative experimental results?
- They are unusable once a model is trained
- They prove the target is invalid
- They inform future model updates and explain rejected designs
- They make every candidate look active