Biopsy samples from brain tumor bank at Columbia University were de-identified and analyzed in alignment with WMA Declaration of Helsinki. Tissue samples were divided, one part H&E stained, the other stored at −80°C. Spatial alignment between biopsy samples and imaging data was ensured using custom registration software and FSL. Three tissue-specific gene modules (Neu, Pro, Inf) were identified through RNAseq-IHC correlation analysis and GSVA. RNA extraction and PLATE-Seq were performed for transcriptome expression. IHC staining and quantification were done for SOX2, Ki67, CD68, and NeuN. MRI sequences included T1Gd, T2, FLAIR, ADC, and SWI. BioNet, a multitask semi-supervised learning model, was constructed to predict gene module expression using both labeled and unlabeled samples, incorporating domain knowledge and UQ.