A semi-supervised weighted SPCA- and convolution KAN-based model for drug response prediction.

Area of Science:

• Genomics
• Computational Biology
• Precision Oncology

Background:

• Predicting drug response in cell lines using multi-omics data is crucial for precision oncology.
• Current methods face challenges in feature extraction, multi-omics data fusion, and handling small sample sizes.
• Overfitting remains a significant concern in deep learning models for this task.

Purpose of the Study:

• To develop an innovative drug response prediction model (NMDP) that addresses the limitations of existing approaches.
• To enhance feature extraction, data fusion, and predictive modeling for multi-omics gene data.
• To improve the accuracy and biological interpretability of drug response predictions.

Main Methods:

• Introduced an interpretable semi-supervised weighted SPCA module for feature extraction from multi-omics gene data.
• Developed a multi-omics data fusion framework utilizing sample similarity networks, bimodal tests, and variance information.
• Combined one-dimensional convolution with Kolmogorov-Arnold Networks (KANs) for drug response prediction.

Main Results:

• The NMDP model achieved superior performance in predicting drug response, with sensitivity and specificity of 0.92 and 0.93, respectively.
• Demonstrated significant performance improvements (11%-57%) compared to seven advanced drug response prediction methods.
• Bio-enrichment experiments validated the biological interpretability and target identification capabilities of the NMDP model.

Conclusions:

• The NMDP model offers a robust and interpretable solution for predicting drug response using multi-omics data.
• The proposed methods for feature extraction and data fusion effectively handle the complexities of multi-omics datasets.
• NMDP shows promise for advancing precision oncology by enabling more accurate drug response predictions and identifying novel therapeutic targets.