Separating drugs from nondrugs: a statistical approach using atom pair distributions.

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Quantifying the "drug-like character" of chemical compounds is crucial for efficient drug discovery.
• Existing methods may be computationally intensive or lack broad applicability.

Purpose of the Study:

• To develop a computationally inexpensive method for quantifying the drug-likeness of chemical compounds.
• To establish a reliable drug-likeness score for filtering large chemical databases.

Main Methods:

• Analysis of atom type and pair-wise combination distributions in known drugs and non-drugs.
• Statistical analysis of occurrence probabilities to derive a logarithmic drug-likeness score.
• No fitting or error minimization schemes were employed.

Main Results:

• A drug-likeness score was derived, with typical pharmaceuticals scoring above 0.3 and ordinary substances below 0.
• The method correctly predicted confirmed drugs with at least 71% accuracy.
• False negatives often exhibited drug-like features or easily identifiable unsuitable patterns.

Conclusions:

• The developed computational approach provides an effective and inexpensive means to assess drug-likeness.
• The drug-likeness score serves as a valuable filter for in silico screening of large substance databases.
• The method complements medicinal knowledge for refining compound selection in drug discovery.