Cancer type and survival prediction based on transcriptomic feature map

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Cancer type and survival prediction remain challenging.
• Transcriptomic data offers insights into cancer biology.
• Current omics analysis methods can be improved for clinical application.

Purpose of the Study:

• To develop a novel pan-cancer transcriptomic feature map for improved cancer type and survival prediction.
• To identify potential cancer biomarkers using deep learning and network analysis.
• To facilitate personalized cancer treatments through advanced omics analysis.

Main Methods:

• Data cleaning, feature extraction, and visualization of TCGA transcriptomic and survival data.
• Construction of a pan-cancer transcriptomic feature map.
• Application of Inception networks and gated convolutional modules for classification.
• Differential gene extraction and interaction network analysis.
• Survival prediction using feature maps and data amplification.

Main Results:

• A pan-cancer transcriptomic feature map was successfully constructed.
• Pan-cancer classification accuracy reached 91.8% using deep learning models.
• Two key genes, ANXA5 and ACTB, were identified as potential biomarkers.
• Survival prediction accuracy ranged from 0.75 to 0.91 for 10 cancer types.

Conclusions:

• The transcriptomic feature map offers a novel approach for cancer omics analysis.
• Identified genes ANXA5 and ACTB show potential as biomarkers for cancer progression and treatment resistance.
• This method can facilitate personalized cancer treatments by reflecting individual differences.