Deep learning-based metabolomics data study of prostate cancer.

Area of Science:

• Biochemistry
• Computational Biology
• Oncology

Background:

• Prostate cancer (PCa) is a heterogeneous disease with diverse clinical and biological features, complicating early diagnosis and treatment.
• Metabolomics offers potential for PCa diagnosis, treatment, and prognosis, but data challenges like high dimensionality and noise hinder classification.
• Deep learning applications in metabolomics research remain underexplored.

Purpose of the Study:

• To develop and evaluate a novel hybrid deep learning model for classifying prostate cancer metabolomics data.
• To enhance biomarker discovery for prostate cancer using advanced computational methods.
• To address the challenges posed by high-dimensional, noisy metabolomics data in cancer research.

Main Methods:

• A hybrid model, TransConvNet, combining transformer and convolutional neural networks, was proposed for PCa metabolomics data classification.
• The model incorporates 1D convolution for attention input, a gating mechanism for attention weight adjustment, and residual networks to mitigate gradient vanishing.
• Comparative experiments with seven machine learning algorithms and validation on a lung cancer dataset were performed. A Mutual Information-based random forest (MI-RF) model was also developed for biomarker identification.

Main Results:

• TransConvNet achieved 81.03% accuracy and an 0.89 AUC in classifying PCa metabolomics data via five-fold cross-validation, outperforming other algorithms.
• The model demonstrated robustness and adaptability by successfully generalizing to a lung cancer dataset.
• The MI-RF model effectively identified key prostate cancer biomarkers, surpassing traditional methods in scope.

Conclusions:

• TransConvNet shows significant potential for accurate classification of prostate cancer metabolomics data.
• The developed models offer valuable tools for biomarker discovery, potentially improving prostate cancer diagnosis and treatment.
• These findings underscore the utility of advanced deep learning and computational approaches in precision oncology and metabolomics research.