Deep learning of 2D-Restructured gene expression representations for improved low-sample therapeutic response...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Clinical outcome prediction is crucial for personalized medicine and stratified therapeutics.
• Machine learning (ML) and deep learning (DL) show promise in predicting therapeutic responses using transcriptomic data.
• Existing DL methods face challenges with low-sample, high-dimensional, and unordered clinical transcriptomic data, limiting prediction accuracy.

Purpose of the Study:

• To develop and evaluate a novel deep learning approach for enhanced clinical outcome prediction from transcriptomic profiles.
• To address the limitations of low-sample sizes and high dimensionality in clinical transcriptomic data analysis.
• To improve the accuracy and robustness of therapeutic response prediction compared to state-of-the-art ML methods.

Main Methods:

• An unsupervised manifold-guided algorithm was used to transform transcriptomic data into ordered, image-like 2D representations.
• Deep convolutional neural networks (ConvNets) were employed for efficient deep learning on these 2D representations.
• The proposed DL models were benchmarked against established ML algorithms on 17 low-sample datasets.

Main Results:

• The developed DL models significantly outperformed state-of-the-art ML models in 82% of the evaluated low-sample benchmark datasets.
• A notable improvement in prediction accuracy (AUC/ACC) exceeding 0.05 was observed in 53% of the datasets.
• The DL models demonstrated superior robustness in cross-cohort prediction tasks and identified reliable biomarkers.

Conclusions:

• The proposed manifold-guided DL approach effectively enhances prediction capabilities for clinical transcriptomic data, especially in low-sample settings.
• This method offers a significant advancement over traditional ML techniques for therapeutic response prediction.
• The findings suggest a promising direction for leveraging deep learning in precision medicine through transcriptomic analysis.