Machine Learning Detects Pan-cancer Ras Pathway Activation in The Cancer Genome Atlas

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Precision oncology currently relies on genomic data to guide cancer treatment, but this approach often misses patients who could benefit from therapy.
• Identifying these
• hidden responders
• is crucial for improving treatment efficacy.
• The tumor transcriptome offers a rich source of molecular information that is currently underutilized in precision oncology.

Purpose of the Study:

• To develop and evaluate a machine-learning (ML) approach for classifying aberrant pathway activity in tumors.
• To identify
• hidden responders
• by analyzing molecular states beyond traditional genomic markers.
• To assess the potential of transcriptomic data combined with ML in precision oncology.

Main Methods:

• An ML algorithm was developed to integrate RNA-sequencing (RNA-seq), copy number, and mutation data.
• The algorithm was applied to 33 cancer types from The Cancer Genome Atlas (TCGA) PanCanAtlas project.
• The model was specifically applied to predict aberrant activity in the Ras signaling pathway and response to MEK inhibitors.

Main Results:

• The ML approach successfully detected Ras pathway activation across various cancer types.
• The method identified phenocopying variants and suggested that multiple genetic alterations in the Ras pathway increase its activity.
• The model, trained on human tumor data, demonstrated the ability to predict MEK inhibitor response in cell lines with wild-type Ras.

Conclusions:

• The tumor transcriptome, when analyzed with machine learning, can reveal aberrant molecular states indicative of potential treatment response.
• This approach holds promise for identifying
• hidden responders
• missed by current genomic-based precision oncology strategies.
• Integrating transcriptomic data and ML can significantly enhance the precision and effectiveness of cancer treatment selection.