Artificial Intelligence in Drug Discovery and Development Nptel Week 3 Answers
Are you looking for AI in Drug Discovery and Development Week 3 Answers ? All weeks solutions of this swayam course assignment are available here.
Table of Contents
AI in Drug Discovery and Development Week 3 Answers (July-Dec 2025)
Course link: Click here to visit course on Nptel Website
Question 1. Statement: Supervised learning always outperforms unsupervised learning in drug discovery applications. Which of the following best evaluates this statement?
a) True; because labeled data in drug discovery is always available.
b) False; because unsupervised learning is essential for tasks like clustering compounds and exploratory analysis.
c) True; since supervised learning can generate models faster than unsupervised ones.
d) False; because unsupervised models don’t require validation or test sets.
Question 2. You are building a classification model for “active” vs “inactive” compounds using SVM. Which of the following would likely reduce model performance?
a) Applying PCA before training
b) Using unnormalized continuous features with very different scales
c) Performing cross-validation
d) Encoding categorical variables using one-hot encoding
Question 3. Statement: Binary features, though limited in range, are valuable for molecular decision-making. This statement is:
a) True, because binary flags help in early decision branching (e.g., Rule of Five compliance).
b) False, since binary features limit model flexibility.
c) False, because neural networks can’t process binary inputs.
d) True, but only in image-based models.
Question 4. Which pair best explains why ReLU is preferred over sigmoid in deep networks?
a) Simpler computation + higher output
b) Better interpretability + zero gradient
c) Non-linearity + reduced vanishing gradient problem
d) Linear behavior + easier convergence
These are AI in Drug Discovery and Development Week 3 Answers
Question 5. You notice your neural network performs well on training but poorly on validation data. What’s the most probable cause?
a) Underfitting due to low learning rate
b) Overfitting due to excessive training epochs or model complexity
c) Using too few activation functions
d) Gradient explosion
Question 6. Statement: Cross-entropy loss is more suitable than MSE for classifying compounds as toxic/non-toxic. This is because:
a) MSE is only used for neural networks, not classification.
b) Cross-entropy penalizes confident but wrong predictions, enhancing classification learning.
c) MSE never works for binary data.
d) Neural networks don’t support MSE.
Question 7. You are training an RNN to predict peptide drug properties from amino acid sequences. What is a key architectural advantage RNNs offer here?
a) High-speed convolutions
b) Backpropagation over batches
c) Memory of sequence order via hidden state
d) Parallel learning of entire input at once
Question 8. Which preprocessing method should be avoided if your dataset contains important extreme ADMET values (outliers)?
a) Min-Max Normalization
b) PCA
c) Log Transformation
d) Mean Imputation
These are AI in Drug Discovery and Development Week 3 Answers
Question 9. Statement: Scaffold-based splitting improves model generalization in chemical datasets. Why?
a) It introduces chemical diversity and reduces overfitting to known chemotypes.
b) It increases model speed by clustering similar molecules.
c) It uses protein-ligand interactions as input features.
d) It avoids removing redundant features.
Question 10. You are handling gene expression data with 20,000 genes. Which approach is best before feeding to ML models?
a) Binary encoding
b) Z-score normalization + differential gene filtering
c) Label encoding + imputation
d) Log transformation + scaffold splitting
Question 11. Why are molecular graphs more useful than fingerprints for GNN models in drug discovery?
a) They simplify the SMILES representation.
b) They preserve spatial and relational information among atoms.
c) They convert binary features into real values.
d) They remove redundant descriptors.
These are AI in Drug Discovery and Development Week 3 Answers
Question 12. A model has 95% accuracy but performs poorly in predicting rare active compounds. What metric should replace accuracy here?
a) RMSE
b) R2 Score
c) F1-Score
d) Silhouette Score
Question 13. Which of the following scenarios best fits using R2 Score as the evaluation metric?
a) Predicting compound classification as “active/inactive”
b) Ranking molecules by efficacy
c) Measuring variance explained in a regression model for IC₅₀ prediction
d) Evaluating how well classes are separated in clustering
These are AI in Drug Discovery and Development Week 3 Answers
Question 14. Your model has high precision but low recall. What does this imply?
a) Most predicted positives are wrong
b) The model misses many actual positive instances
c) The model has overfitting due to hyperparameters
d) The test data is incorrect
Question 15. You want to visualize a protein-ligand interaction inside a Jupyter notebook. What should you use?
a) Seaborn
b) DeepChem
c) Py3Dmol or NGLView
d) NumPy
These are AI in Drug Discovery and Development Week 3 Answers