Abstract
The TRIM CIV subfamily, distinguished by its C-terminal PRY-SPRY domains, constitutes nearly half of the human TRIM family and plays pivotal roles in cancer progression through ubiquitination. Identifying TRIM CIV substrates and interactors has emerged as a critical approach for elucidating tumorigenesis. Current protein-protein interaction (PPI) prediction models face challenges, including an inherent deficiency of negative datasets, biased feature integration, and the absence of a cancer-specific interaction context. To achieve the precise identification of TRIMCIV targets, we developed TRIMCIVtargeter with predictive models that systematically integrates multi-dimensional PPI features-expression differences and correlations in specific cancer, comparable protein-docking scores, and cancer-specific context. Learning from the functional and structural interaction features between 718 experimentally validated TRIM-target pairs, two types of SVM-based binary models were independently trained using proteomic and transcriptomic data. Our models achieved robust prediction performance in cancers utilizing a fair feature space and circumventing hypothetical non-interacting pairs. TRIMCIVtargeter not only provides a cancer-related resource for studying TRIMCIV-mediated regulatory mechanisms but also offers a new perspective for family-specific PPI prediction, holding significant implications for biomarker discovery and therapeutic targeting in oncology. The online platform of TRIMCIVtargeter is now available.