Investigating the Behavior of Published PAINS Alerts Using a Pharmaceutical Company Data Set

Area of Science:

• Medicinal Chemistry
• Drug Discovery
• Biochemistry

Background:

• Biochemical assay interference is a major resource drain in drug discovery.
• Pan-assay interference compounds (PAINS) alerts were previously identified to flag such issues.
• The original PAINS alert dataset was limited, necessitating further validation.

Purpose of the Study:

• To rigorously assess the validity of PAINS alerts using a large, internal drug discovery dataset.
• To evaluate PAINS alerts based on promiscuity across multiple assay formats, compound stability, cytotoxicity, and binding characteristics.
• To determine if PAINS alerts effectively identify compounds causing assay interference.

Main Methods:

• Analysis of a large internal dataset from Lilly.
• Evaluation of PAINS alerts across four criteria: promiscuity (including AlphaScreen), compound stability, cytotoxicity, and high Hill slope (as a surrogate for non-specific binding).
• Statistical assessment of alert performance in predicting interference phenomena.

Main Results:

• Only three PAINS alerts demonstrated significant pan-assay promiscuity.
• The majority of PAINS alerts primarily encoded molecules promiscuous in AlphaScreen assays.
• Many PAINS alerts identified compounds exhibiting instability, cytotoxicity, or non-specific binding (high Hill slope).

Conclusions:

• PAINS alerts are not consistently predictive of pan-assay promiscuity across diverse formats.
• PAINS alerts are more indicative of issues in specific assay technologies like AlphaScreen.
• While not universally predictive of promiscuity, PAINS alerts remain valuable for identifying problematic compounds related to stability, toxicity, and binding in drug discovery pipelines.