Artificial intelligence uncovers carcinogenic human metabolites.

Area of Science:

• Genomics
• Toxicology
• Computational Biology

Background:

• Eukaryotic genomes face constant threats from various compounds, potentially leading to malignant transformation despite DNA repair mechanisms.
• Accurate prediction of carcinogens remains challenging due to limited data on known carcinogens and non-carcinogens.

Purpose of the Study:

• To develop an accurate and interpretable carcinogen prediction tool named Metabokiller.
• To identify potential carcinogenic human metabolites using the developed classifier.

Main Methods:

• Developed Metabokiller, an ensemble classifier assessing electrophilicity, proliferation, oxidative stress, genomic instability, epigenome alterations, and anti-apoptotic responses.
• Validated predictions using functional assays in Saccharomyces cerevisiae and human cells with flagged metabolites (4-nitrocatechol, 3,4-dihydroxyphenylacetic acid).

Main Results:

• Metabokiller accurately recognizes carcinogens and outperforms existing prediction methods.
• The tool identified potential carcinogenic human metabolites.
• Experimental validations showed high synergy with Metabokiller predictions for flagged metabolites.

Conclusions:

• Metabokiller offers a robust and interpretable approach to carcinogen prediction.
• The study highlights potential risks associated with specific human metabolites.
• The findings pave the way for improved toxicological assessments and cancer prevention strategies.