BMPCD: A Pan-Cancer Detection Method Base on Learning Cross-Domain Feature

The microbiome has emerged as a predictor of several diseases.Increasing evidence suggests that the tumor microbiome is involved in cancer growth.Additionally, host information like physiological characteristics can alter the structure of the microbiome, but this information has not yet been fully utilized in data models.Language representation models have gained widespread attention in the field of natural language processing (NLP), owing to their remarkable effectiveness.In this study, we propose a novel approach that uses the BERT model to extract features from textual host information and then incorporates them into microbiome-based modeling.The experimental results showed that when only the microbiome was used, the AUROC values were lower, but the performance of the cancer screening model could be significantly improved with embedding host information.