Abstract 370P
Background
In lung adenocarcinoma, therapies like EGFR TKIs and ICIs have improved outcomes for EGFR-mutant and wild-type patients, respectively. However, drug resistance and limited survival remain significant challenges. This study proposes a novel approach using whole transcriptome data to develop a prognosis prediction model through an image-based deep learning method.
Methods
Four cohorts were analyzed: one training cohort (RNA-seq, n=391) and three independent validation cohorts (TCGA RNA-seq, n=394; GSE68465, n=443; GSE13213, n=117). RNA-seq data were transformed into images using pixel encoding strategies designed to preserve transcriptomic information. The Convolutional Neural Network (CNN) model was trained on all gene expression data without feature selection. To prevent overfitting, the training cohort was divided into subsets: training (n=259), testing (n=62), and validation (n=77). The CNN model was trained on the training set, validated on the testing and validation sets, and independently evaluated on the three validation cohorts.
Results
RNA-seq data were successfully transformed into images first. The image transformation process enabled effective training of the CNN model, which categorized patients into high- and low-risk groups. In the training cohort, high-risk patients had significantly shorter overall survival (all P < 0.01). Similar results were observed in external validation cohorts: TCGA (P=0.001), GSE68465 (P < 0.0001), and GSE13213 (P=0.003). Prediction accuracies were 0.71, 0.81, and 0.92 for the training, testing, and validation subsets, respectively. In the external cohorts, prediction accuracy averaged 0.7.
Conclusions
This study introduces an innovative image-based deep learning strategy to analyze whole transcriptome data without requiring differential gene selection. This approach captures comprehensive transcriptomic information, offering potential for improved prognostic modeling and molecular guidance in lung cancer treatments.
Editorial acknowledgement
During the preparation of this work the author(s) used ChatGPT in order to do the English editing of this abstract. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.
Funding
Academia Sinica and National Science and Technology Council (Taiwan)
Disclosure
All authors have declared no conflicts of interest.