Conformal Selection for Efficient and Accurate Compound Screening in Drug Discovery.

Area of Science:

• Drug discovery and development
• Computational chemistry
• Statistical inference

Background:

• Compound screening is crucial for drug discovery but suffers from inaccurate false compound selection or omission.
• Current methods lack robust risk controls, hindering the identification of effective drug candidates.

Purpose of the Study:

• To introduce conformal selection, an enhanced approach for optimizing compound screening.
• To provide rigorous control over false discovery and omission rates in molecule selection.
• To improve the power of compound screening for identifying promising drug candidates.

Main Methods:

• Leveraging conformal inference to generate p-values for each candidate molecule.
• Quantifying statistical evidence for molecule selection using p-values.
• Comparing p-values against thresholds derived from multiple testing principles for final selection.

Main Results:

• The conformal selection approach offers rigorous control over false discovery/omission rates.
• The method's validity is independent of dataset size and requires minimal assumptions.
• Conformal selection achieves higher statistical power by avoiding prediction error estimation, enhancing candidate identification.

Conclusions:

• Conformal selection provides a statistically rigorous and powerful method for optimizing compound screening in drug discovery.
• This approach balances risks and benefits, improving the reliability of identifying promising drug candidates.
• The method's independence from dataset size and minimal assumptions make it broadly applicable.