Development and experimental test of support vector machines virtual screening method for searching Src inhibitors...

Area of Science:

• Biochemistry and Molecular Biology
• Computational Chemistry
• Pharmacology

Background:

• Src kinase is a key target in cancer progression, invasion, and metastasis.
• Drug resistance necessitates the discovery of novel Src inhibitors.
• In-silico methods are crucial for efficient screening of potential inhibitors.

Purpose of the Study:

• To evaluate Support Vector Machines (SVM) as a virtual screening tool for identifying Src inhibitors.
• To assess SVM's performance in searching large compound libraries for potential Src inhibitors.
• To compare SVM with other machine-learning and similarity-based virtual screening methods.

Main Methods:

• Trained and tested SVM models using a dataset of 1,703 known Src inhibitors and 63,318 non-inhibitors.
• Performed 5-fold cross-validation to assess model accuracy.
• Screened large compound databases (PubChem, MDDR) using the trained SVM models.

Main Results:

• SVM achieved high accuracy, correctly identifying 93.53%–95.01% of inhibitors and 99.81%–99.90% of non-inhibitors in cross-validation.
• SVM successfully identified 70.45% of newly reported Src inhibitors (since 2011).
• SVM predicted a low percentage of false positives when screening large compound libraries.

Conclusions:

• SVM demonstrates comparable yield and reduced false-hit rates compared to other virtual screening methods.
• SVM shows potential for efficiently searching large compound libraries for Src inhibitors.
• Experimental validation of SVM-identified compounds confirmed moderate activity for one novel scaffold, supporting SVM's utility.