This invention discloses a method and
system for
cancer prognostic analysis based on multi-
omics data, belonging to the field of
data analysis. This invention employs a variational
autoencoder with an attention-integrated mechanism to achieve
nonlinear dimensionality reduction, importance weighting, and deep encoding of high-dimensional multi-
omics features. The decoder reconstructs input features from latent variables to retain original information. The
loss function integrates reconstruction loss and KL
divergence, and adjustable weights balance feature reconstruction accuracy and distribution constraints. This model retains biological
interpretability while extracting features through
deep learning, providing high-quality deep features for subsequent prognostic modeling. Through the organic combination of
bioinformatics analysis and
deep learning models, it achieves high-accuracy, robust, and highly interpretable prognostic predictions in pan-
cancer scenarios, effectively addressing the technical shortcomings of traditional pan-
cancer prognostic analysis, such as insufficient Cox model capabilities,
deep learning black box issues, model
overfitting, and lack of an integrated
system.