DeeP4med: deep learning for P4 medicine to predict normal and cancer transcriptome in multiple human tissues

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• P4 medicine (predict, prevent, personalize, participate) emphasizes individualized disease diagnosis and prediction.
• Gene expression data analysis is crucial for predicting disease states.
• Deep learning models offer intelligent strategies for disease prediction using gene expression profiles.

Purpose of the Study:

• To develop a deep learning model for predicting cancer gene expression (mRNA) matrices from matched normal samples and vice versa.
• To evaluate the model's performance in tissue and disease classification.
• To compare the model's efficacy against traditional machine learning algorithms.

Main Methods:

• An autoencoder deep learning model, DeeP4med, was designed, incorporating a Classifier and a Transferor.
• The model was trained to predict gene expression matrices between normal and cancer tissues.
• Performance was assessed using F1 scores and accuracy for classification tasks.

Main Results:

• DeeP4med achieved high F1 scores, ranging from 0.935 to 0.999, depending on tissue type.
• The model demonstrated excellent accuracy: 0.986 for tissue classification and 0.992 for disease classification.
• DeeP4med outperformed seven classic machine learning models in predictive accuracy.

Conclusions:

• The DeeP4med model can predict tumor gene expression from normal tissue, facilitating the identification of key genes involved in cancer development.
• Analysis of predicted matrices showed strong correlation with existing literature and databases for 13 cancer types.
• This approach enables prediction of cancer diagnosis from healthy tissue gene expression and suggests potential therapeutic interventions.