HONeYBEE is an open-source multimodal biomedical data integration framework for oncology applications. It processes structured and unstructured clinical data, whole-slide images, radiology scans, and molecular profiles, generating integrated patient-level embeddings using domain-specific base models and fusion strategies. These embeddings enable survival prediction, cancer type classification, patient similarity retrieval, and cohort clustering. When evaluated on over 11,400 patients across 33 cancer types from the TCGA, the clinical embeddings demonstrated the strongest unimodal performance, with 98.5% classification accuracy and 96.4% precision @10 for patient retrieval. It also achieved the highest survival prediction concordance index across most cancer types. Multimodal fusion offers complementary benefits for specific cancers, improving overall survival prediction beyond what clinical features alone can achieve. Comparative evaluations of four large-scale language models show that general-purpose models like Qwen3 improve task-specific fine-tuning performance on heterogeneous data such as pathology reports, but outperform specialized medical models on clinical text representation.
