Identification of pan-cancer Ras pathway activation with deep learning

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Precision oncology aims to identify patient subgroups who will respond to specific therapies.
• Current machine learning models for classifying aberrant pathway activity from multiomic cancer data face limitations including high-dimensionality and data sparsity.
• Accurate identification of hidden responders is crucial for effective cancer treatment strategies.

Purpose of the Study:

• To develop a robust deep learning model for detecting Ras pathway activation across diverse cancer types.
• To address the limitations of existing methods in identifying hidden responders in precision oncology.
• To improve the diagnosis of RAS aberrant pathway activity using multiomic data.

Main Methods:

• Development of the Nature-inspired deep Ras activation pan-cancer (NatDRAP) model, a deep neural network (DNN).
• Integration of bulk RNA sequencing, copy number, and mutation data from PanCanAtlas.
• Synergistic optimization of DNNs using a nature-inspired artificial bee colony algorithm combined with gradient-based optimizers.
• Validation across 33 different cancer types.

Main Results:

• NatDRAP demonstrates superior performance and robustness in diagnosing RAS aberrant pathway activity compared to benchmark methods.
• The model effectively detects pan-cancer Ras pathway activation using integrated multiomic data.
• Gene ontology enrichment and pathological analysis provide novel insights into RAS pathway activity.

Conclusions:

• NatDRAP offers a powerful and robust solution for identifying hidden responders in precision oncology by accurately detecting Ras pathway activation.
• The proposed deep learning framework overcomes limitations of previous approaches in handling high-dimensional and sparse multiomic data.
• This work facilitates improved characterization and diagnosis of RAS aberrant pathway activity, advancing personalized cancer treatment.