Discovery of novel SOS1 inhibitors using machine learning

Area of Science:

• Oncology
• Pharmacology
• Computational Chemistry

Background:

• RAS signaling pathway overactivation drives 30% of human cancers.
• Son of sevenless 1 (SOS1) is a key regulator of RAS activation, making it a therapeutic target for RAS-driven malignancies.

Purpose of the Study:

• To identify novel small-molecule inhibitors of SOS1 using machine learning-based virtual screening.
• To explore new chemical scaffolds for SOS1 inhibition in cancer therapy.

Main Methods:

• Utilized a random forest regressor for virtual screening of large compound libraries (L-series, EGFR-related, and CNCL).
• Evaluated inhibitory activity using KRAS G12C/SOS1 protein-protein interaction (PPI) assays.
• Performed molecular docking and in silico drug-likeness assessments.

Main Results:

• Discovered nine novel compounds with unexplored chemical frameworks exhibiting SOS1 inhibitory activity.
• The most potent compound achieved >50% inhibition and an IC50 near 20 μg mL−1.
• Hit compounds, like CL01545365, bind to a unique pocket within the target, distinct from known inhibitors.

Conclusions:

• Novel SOS1 inhibitors with unique binding modes were identified from the carboxylic acid series.
• These compounds represent promising candidates for developing new therapeutics against RAS-driven cancers.
• The findings highlight the potential of ML-driven VS in discovering novel drug scaffolds.