Currently, many studies on neuropsychiatric disorders have utilized massive trio-based whole-exome sequencing (WES) and whole-genome sequencing (WGS) to identify numerous de novo mutations (DNMs). Here, we retrieved 17,104 DNMs from 3,555 trios across four neuropsychiatric disorders: autism spectrum disorder (ASD), epileptic encephalopathy (EE), intellectual disability (ID), schizophrenia (SCZ), in addition to unaffected siblings (Control), from 36 studies by WES/WGS. After eliminating non-exonic variants, we focused on 3,334 exonic DNMs for evaluation their association with these diseases. Our results revealed a higher prevalence of DNMs in the probands of all four disorders than the one in the controls (P < 1.3 × 10-7). The elevated DNM frequency is dominated by loss-of-function/deleterious single nucleotide variants and frameshift indels (i.e., extreme mutations, P < 4.5 × 10-5). With extensive annotation of these “extreme” mutations, we prioritized 764 candidate genes in these four disorders. A combined analysis of Gene Ontology, microRNA targets, and transcription factor targets revealed shared biological process and non-coding regulatory elements of candidate genes in the pathology of neuropsychiatric disorders. In addition, weighted gene co-expression network analysis (WGCNA) of human laminar-specific neocortical expression data showed that candidate genes are convergent on eight shared modules with specific layer-enrichment and biological process features. Furthermore, we identified that 53 candidate genes are associated with more than one disorder (P < 0.000001), suggesting a possibly shared genetic etiology underlying these disorders. Particularly, DNMs of the SCN2A gene are frequently occurred across all four disorders. Finally, we constructed a freely available NPdenovo database, which provides a comprehensive catalog of the DNMs identified in neuropsychiatric disorders.