引導式測驗 · 中等 等級
AI酵素工程測驗
應用機器學習輔助的定向演化概念,同時考慮適應性景觀、上位性、分析背景、不確定性和實驗驗證。
問題 1 的 8
What data does an enzyme fitness model learn from?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- What data does an enzyme fitness model learn from?
- Unlabeled sequence records with no target property measurements
- A list of candidate sequences without assay outcomes
- Protein variants paired with assay-defined measurements
- Structural coordinates without the activity endpoint being modeled
- How can combined mutations behave on a protein fitness landscape?
- Each mutation contributes the same effect in every sequence background
- Mutation effects can depend on which other mutations are present
- Combining substitutions cannot change the effect of an individual mutation
- An additive score captures every interaction between substitutions
- Why use uncertainty when selecting the next batch of variants?
- It identifies the highest true activity without another experiment
- It makes screening measurements unnecessary
- It separates measurement noise from biological signal perfectly
- It can identify candidates that are promising or informative to test
- Why include assay conditions with sequence measurements?
- Activity depends on temperature, substrate, pH and other experimental context
- An assay label makes results comparable even when conditions differ
- Recording conditions guarantees transfer to a new process
- The same variant has an invariant measured score across assay environments
- What can happen when a model trained on single mutations predicts combinations?
- Each combined variant inherits the sum of single-mutation effects
- The assay can no longer distinguish activity between variants
- Mutation interactions can make combined effects non-additive
- Single-mutation measurements fully determine every combination
- Which evaluation best tests prospective enzyme engineering performance?
- Evaluate only the model fit on the same variants used for training
- Measure selected variants experimentally in a later or held-out round
- Include near-identical copies of each candidate in both partitions
- Use test outcomes to choose the candidates before evaluation
- Why compare ML-guided selection with a simple baseline?
- To establish that the simple method always finds the global optimum
- To avoid collecting repeat measurements for noisy assays
- To make a model and baseline use different target measurements
- To determine whether model-guided choices add value
- What validates an engineered enzyme for an intended process?
- Experimental testing under relevant process conditions
- A high prediction from a screening assay under unrelated conditions
- A close match between the candidate embedding and a training embedding
- A sequence annotation without measurements of the intended property