Bài kiểm tra có liên kết hướng dẫn · Cứng Cấp độ
Câu đố về Hóa học Sáng tạo và Thiết kế Phân tử De Novo
Hiểu cách hóa học tổng hợp đề xuất ứng viên và đánh giá giá trị, mục tiêu, tính mới, khả năng tổng hợp và theo dõi thử nghiệm của hóa học.
Câu hỏi 1 của 8
What does de novo molecular design ask a model to do?
Xem trước mọi câu hỏi8 câu hỏi
Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.
- What does de novo molecular design ask a model to do?
- Only classify known compounds
- Measure a clinical endpoint directly
- Propose new molecular structures under stated objectives
- Replace chemical synthesis
- Why can a model exploit a molecular reward function?
- It always learns the true biological mechanism
- It may find structures that score well because the predictor has blind spots
- Reward scores are direct experimental measurements
- Chemical constraints guarantee activity
- What does a valid molecular graph establish?
- That the compound binds the target
- That it can be synthesized at scale
- That clinical efficacy is proven
- That the structure follows encoding rules, not that it is useful or safe
- How should novelty be interpreted in a generative chemistry report?
- Relative to a specified and dated reference set after standardization
- As proof that a molecule has never existed anywhere
- Only by counting generated strings
- Without considering salts or stereochemistry
- Why include synthesis feasibility review?
- Models can rank commercially available reagents without route planning
- Synthesis feasibility proves the predicted target activity
- A predicted property score does not show that a route is practical
- A high reward guarantees the structure can be made at scale
- What can go wrong when optimizing only a predicted activity score?
- It guarantees molecular diversity
- The predictor may be wrong outside its training domain
- It prevents reward hacking
- It measures the real assay directly
- Which evaluation set design better tests chemical generalization?
- Random duplicate structures across every split
- The training set alone
- Only candidates with the highest reward
- A scaffold-separated held-out set
- What ultimately establishes whether a proposed candidate has the desired biological behavior?
- Experimental assays and expert interpretation
- A validity filter alone
- The model architecture name
- A high generative likelihood