Highly predictive support vector machine (SVM) models for anthrax toxin lethal factor (LF) inhibitors

Area of Science:

• Biochemistry
• Computational Biology
• Toxicology

Background:

• Anthrax is a lethal disease caused by *Bacillus anthracis*, with the lethal factor (LF) toxin being a primary driver of pathogenesis and host death.
• Current antibiotics clear bacteria but do not neutralize LF toxin, highlighting the need for effective LF inhibitors.
• Existing LF inhibitor scaffolds lack sufficient efficacy, selectivity, and safety for therapeutic use.

Purpose of the Study:

• To develop accurate computational models for rapid identification of novel, structurally diverse chemical inhibitors of anthrax lethal factor (LF).
• To address the unmet therapeutic need for agents that can neutralize LF-mediated toxemia.

Main Methods:

• Development and validation of support vector machine (SVM) models using published anthrax lethal factor (LF) biological activity data.
• Training and testing involved 508 active compounds and 847 inactive compounds from the PubChem BioAssay database.
• External validation of model M1 against unbiased test sets to assess prediction accuracy and selectivity.

Main Results:

• The developed SVM models demonstrated high prediction accuracy for LF inhibitory activity.
• Model M1 achieved excellent selectivity, correctly identifying 95.12% of nanomolar LF inhibitors and 93.88% of inactives in external test sets.
• Model M1 showed high accuracy in classifying inactive compounds, predicting 99.65% of PubChem inactives correctly.

Conclusions:

• Novel SVM models provide a powerful tool for predicting LF inhibitory activity and identifying potential novel LF inhibitors from large compound libraries.
• These computational models are expected to accelerate the discovery of much-needed therapeutics for anthrax toxin neutralization.
• The developed models offer a promising approach to overcome limitations of current LF inhibitor scaffolds.