Diffusion kernel-based predictive modeling of KRAS dependency in KRAS wild type cancer cell lines.

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• RAS proteins are key in cell signaling, even without mutations.
• Predicting tumor dependency on RAS is crucial for targeted therapies.
• KRAS inhibitors are advancing, but their application beyond mutated KRAS is unclear.

Purpose of the Study:

• To identify cancer cell lines dependent on KRAS expression, irrespective of mutational status.
• To develop predictive models for KRAS dependency using machine learning and transcriptomic data.
• To explore therapeutic strategies for KRAS wild-type cancers.

Main Methods:

• Analysis of CRISPR/RNAi data from over 700 cancer cell lines.
• Machine learning modeling combined with whole transcriptome data.
• Variable selection using protein-protein interaction network analysis and diffusion kernels.

Main Results:

• Identified a subset of KRAS wild-type (KRASwt) cell lines dependent on KRAS expression.
• Developed a model that accurately predicts KRAS dependency in KRASwt and all cell lines.
• Previous methods based on RAS activating events or RNA signatures were less reliable.

Conclusions:

• KRAS dependency exists in KRAS wild-type cancers, presenting a therapeutic vulnerability.
• Machine learning and transcriptomic analysis can identify these KRAS-dependent tumors.
• This approach may guide targeted therapies for KRAS wild-type cancers.