Super.FELT: supervised feature extraction learning using triplet loss for drug response prediction with multi-omics...

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Precision oncology relies on accurate prediction of patient drug response.
• Multi-omics data improve drug response prediction but present dimensionality challenges.
• Developing methods to handle high-dimensional multi-omics data is crucial for clinical applications.

Purpose of the Study:

• To develop a novel supervised deep learning method for drug response prediction.
• To effectively reduce the dimensionality of multi-omics data for improved prediction accuracy.
• To validate the proposed method on diverse datasets including cell lines, patient-derived tumor xenografts, and The Cancer Genome Atlas data.

Main Methods:

• Proposed Supervised Feature Extraction Learning using Triplet loss (Super.FELT).
• Employed a three-stage approach: feature selection, supervised feature encoding, and binary classification.
• Utilized multi-omics data: mutation, copy number aberration, and gene expression from GDSC, CCLE, CTRP, PDX, and TCGA datasets.

Main Results:

• Super.FELT demonstrated superior performance in external validation on PDX and TCGA datasets.
• The method showed good performance in cross-validation on GDSC and external validation on CCLE and CTRP.
• Ablation studies confirmed the benefit of multi-omics data and the three-stage approach for drug response prediction.

Conclusions:

• Super.FELT achieved better performance by separating feature extraction and classification stages.
• Independent training of encoders and classifiers is vital, particularly for external validation on non-cell line data (PDX, TCGA).
• Multi-omics data offer superior performance for external validation compared to gene expression alone, especially for non-cell line datasets.