Most Influential Physicochemical and In Vitro Assay Descriptors for Hepatotoxicity and Nephrotoxicity Prediction

Area of Science:

• Drug discovery and development
• Toxicology
• Computational chemistry

Background:

• Drug-induced organ injury is a significant cause of drug candidate failure.
• The liver, kidneys, and heart are most frequently affected by drug toxicity.
• Current methods like in silico and in vitro assays are used separately to predict drug safety.

Purpose of the Study:

• To combine physicochemical properties and in vitro cytotoxicity assays to predict clinical organ toxicity.
• To develop organ-specific predictive models for hepatotoxicity, nephrotoxicity, and cardiotoxicity.
• To utilize logistic regression models with odds ratios for toxicity prediction.

Main Methods:

• Utilized physicochemical property calculations and in vitro cytotoxicity assays (including mitochondrial dysfunction).
• Developed organ-specific univariate and multivariable logistic regression models.
• Analyzed 215 marketed drugs to establish predictive models for liver, kidney, and heart toxicity.

Main Results:

• Multivariable models predicted hepatotoxicity based on mitochondrial inhibition (OR 6.2) or high plasma concentration (OR 7.5).
• Nephrotoxicity was predicted by high plasma concentration (OR 5.8), mitochondrial inhibition (OR 6.4), or hydrogen-bond-acceptor atoms (OR 15.9).
• Reduced kidney injury risk was associated with polar surface area ≥75 Å (OR 0.21) and ECCS class 2 drugs (OR 0.23); ECCS class 4 drugs showed higher risk (OR 4). Cardiotoxicity model was poorly defined.

Conclusions:

• Combining in silico physicochemical properties and in vitro toxicity assays can build predictive models for drug safety.
• These models can aid in selecting small molecule therapeutics with lower potential for liver and kidney organ toxicity.
• The predictive models show promise for improving early-stage drug safety assessments.