Structural Similarity, Activity, and Toxicity of Mycotoxins: Combining Insights from Unsupervised and Supervised...

Area of Science:

• Computational toxicology
• Cheminformatics
• Mycology

Background:

• Many mycotoxins lack toxicological assessment.
• Current methods overlook co-occurring compounds and diverse fungal metabolites.
• Need for efficient, broad-spectrum toxicity prediction.

Purpose of the Study:

• To develop an in silico method for predicting mycotoxin toxicity.
• To assess molecular similarity and toxicity using defined molecular descriptors.
• To enable rapid, nontarget screening of multiple mycotoxin classes.

Main Methods:

• Unsupervised and supervised machine learning (Random Forest, Neural Networks).
• Integration of molecular and physicochemical descriptors.
• Clustering analyses for structural similarity and toxicity prediction.

Main Results:

• Clustering successfully grouped mycotoxins by family, indicating shared molecular properties.
• Topological descriptors were key in differentiating toxic from non-toxic compounds.
• Machine learning models (RF, NN) improved toxicity prediction accuracy.

Conclusions:

• In silico prediction of mycotoxin toxicity is feasible using molecular descriptors and machine learning.
• A combination of structural and biological activity descriptors is crucial for accurate toxicity profiling.
• This approach facilitates broader toxicological evaluations of fungal metabolites.