Trade-off between accuracy and interpretability for predictive in silico modeling.

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning

Background:

• Investigating the trade-off between model accuracy and interpretability in predictive in silico modeling.
• Accurate models are often complex and opaque, while interpretable models may lack predictive power.

Purpose of the Study:

• To compare state-of-the-art methods for generating accurate predictive models with those for generating transparent models.
• To evaluate the performance penalty associated with choosing interpretable models over opaque ones.

Main Methods:

• Comparison of state-of-the-art accurate modeling techniques against state-of-the-art transparent modeling techniques.
• Evaluation across 16 distinct biopharmaceutical classification tasks.

Main Results:

• Opaque methods generally achieved higher accuracy than transparent methods.
• The performance penalty for selecting interpretable models was often limited.

Conclusions:

• Interpretable models provide a viable alternative when a balance between accuracy and understanding is desired.
• Limited predictive performance loss supports the use of interpretable models in biopharmaceutical classification.