Identifying molecular drivers of clinically aggressive phenotypes in genetic diseases, particularly neurocognitive disorders (NDs), is challenging. Current models often miss key genes, leaving over 50% of individuals undiagnosed and untreated due to high rates of Variants of Uncertain Significance (VUS). We introduce CLinNET, a multi-modal deep neural network designed to improve VUS interpretation and biomarker discovery by integrating sequencing data with tissue-specific gene expression, biological pathways, and gene ontology terms. CLinNET’s architecture leverages sparse networks enriched with Reactome pathways and integrates gene ontology terms, targeting relevant tissues for precise prediction. By employing a novel lyerwise SHapley Additive exPlanations (SHAP) for interpretability, CLinNET provides insights into molecular distinctions between healthy and affected samples, supporting diagnostic decisions.Trained and validated on a ND dataset of 47,328 cases and controls, CLinNET achieved an AUC of 87% and outperformed current models by identifying significantly more ND-associated genes (p-value < 0.001), with minimal overlap with non-ND conditions (p-value = 0.8), underscoring its disease-specific precision. Biological interpretation revealed novel gene ontology terms and pathways relevant to neurocognitive function, such as glutamatergic synapse organization and regulation of trans-synaptic signaling. Among the top 10% genes ranked by CLinNET, 376 were brain-enriched, and 176 linked to nervous system phenotypes when mutated in model organisms—an enrichment not seen in random gene sets (p-value < 0.00001). Notably, 478 genes have known associations with rare diseases, indicating their diagnostic potential. Protein-protein interaction analysis confirmed that 86% of candidate genes are directly interconnected (p-value < 0.001), suggesting shared pathways critical to neurocognitive function.To our knowledge,CLinNET is the first supervised model for ND gene curation, advancing interpretability and uncertainty quantification. This model has transformative potential for early identification and targeted interventions for genetic diseases.