All

Question 1

Is SMILES always unique for each molecule?

Answer 1

False. SMILES is not guaranteed to be unique unless canonicalized.

Question 2

What are molecular fingerprints?

Answer 2

Binary or numerical representations of molecular structures.

Question 3

What are graph representations of molecules in GNNs?

Answer 3

Atoms as nodes and bonds as edges.

Question 4

Do degree matrices contain node features like atom type?

Answer 4

False. Degree matrices contain node degrees, not chemical features.

Question 5

Can autoregressive models be used to generate SMILES?

Answer 5

True. They can sequentially generate valid SMILES strings.

Question 6

What is QSAR modeling used for?

Answer 6

Predicting properties like toxicity, bioactivity, or solubility from structure.

Question 7

Do QSAR datasets usually have millions of data points?

Answer 7

False. They are often small, hundreds to thousands of points.

Question 8

Is cross-validation necessary in QSAR model evaluation?

Answer 8

True. It’s important to avoid overfitting and estimate generalization.

Question 9

Does hyperparameter selection affect model generalizability?

Answer 9

True. It strongly impacts performance and overfitting risk.

Question 10

What are bioassays?

Answer 10

Wet-lab procedures to evaluate compound effects in biological systems.

Question 11

What is drug synergy?

Answer 11

When two drugs combined have a stronger effect than the sum of individual effects.

Question 12

What is target deconvolution?

Answer 12

Predicting all potential protein targets a molecule may bind.

Question 13

What is ligand-based virtual screening?

Answer 13

Finding similar molecules to known actives.

Question 14

Are virtual screening tools 100% accurate?

Answer 14

False. Their predictions are approximations and can be noisy.

Question 15

What are limitations of GNNs in molecular modeling?

Answer 15

They may struggle with very large or complex graphs.

Question 16

What does FCD (Frechet ChemNet Distance) measure?

Answer 16

Distance between feature distributions, not individual atoms.

Question 17

What is synthesis planning?

Answer 17

Designing a sequence of reactions to produce a target compound.

Question 18

What are reaction templates?

Answer 18

Transformation rules used to model chemical reactions.

Question 19

Is chemical equilibrium indicated by a one-way arrow?

Answer 19

False. Equilibrium implies two-way reaction.

Question 20

What are actions in retrosynthesis game?

Answer 20

Choosing reactants to break down a compound.

Question 21

What are challenges in deep learning for medical imaging?

Answer 21

Image quality and acquisition variability.

Question 22

Does U-Net use a symmetrical encoder-decoder structure?

Answer 22

True. It’s a key feature of U-Net.

Question 23

What does PPV (Positive Predictive Value) indicate?

Answer 23

Probability that a positive prediction is truly positive.

Question 24

What are batch effects?

Answer 24

Artificial variation caused by experimental conditions.

Question 25

Is 'smoking status' a continuous variable?

Answer 25

False. It's a categorical variable.

Question 26

Can domain shifts affect the input distribution p(x)?

Answer 26

True. Covariate shift affects p(x).

Question 27

What does Grad-CAM do?

Answer 27

Highlights image regions important for a model’s prediction.

Question 28

Does interpretability equal causality?

Answer 28

False. Interpretable features may still be non-causal.

Question 29

Can MIL handle bags with variable numbers of elements?

Answer 29

True. MIL is designed for that.

Question 30

Is multimodal contrastive learning supervised?

Answer 30

False. It's typically unsupervised.

Question 31

What are DNA sequences composed of?

Answer 31

Nucleotides.

Question 32

Does AlphaFold 2 use CNNs?

Answer 32

False. AlphaFold 2 uses transformers, not CNNs.

Question 33

Can ML predict protein structures from sequences?

Answer 33

True. It's a key use case in bioinformatics.

Question 34

What does ultrasound imaging use?

Answer 34

High-frequency sound waves.

Question 35

Why is domain knowledge important in healthcare ML?

Answer 35

It helps interpret features and avoid misleading correlations.