Paper Title
“Enhancing Kabuki Syndrome Diagnosis with Generative Ai Techniques”

Abstract
Kabuki Syndrome (KS) is a complex, multisystemic genetic disorder often presenting with a broad range of symptoms, including distinct facial dysmorphisms, intellectual disability, skeletal abnormalities, and immunological deficiencies. The diagnosis of KS poses a significant challenge due to its phenotypic variability and overlap with other syndromes. Traditional diagnostic methods rely heavily on expert clinical evaluation and genetic testing, which can be both time-consuming and costly. In this study, we propose a novel approach to enhance KS diagnosis by leveraging generative AI models, specifically designed for the nuanced interpretation of clinical phenotypic data. We introduce phenoBCBERT, a BERT-based model trained on phenotype-specific data, and PhenoGPT, a generative language model, to systematically capture and interpret complex KS-related phenotypes from clinical narratives. phenoBCBERT is optimized for phenotype term extraction and mapping, enabling it to effectively identify relevant phenotypic descriptors directly from unstructured clinical texts. Meanwhile, PhenoGPT synthesizes these descriptors to generate comprehensive, case-specific diagnostic suggestions, cross-referencing patient characteristics with known KS phenotypes. Through extensive evaluation on clinical datasets, our approach demonstrates significant improvements in diagnostic sensitivity and specificity, particularly in distinguishing KS from phenotypically similar conditions. This hybrid AI framework not only expedites the diagnostic process but also reduces the reliance on costly genetic testing by providing a robust preliminary assessment. Additionally, our models offer interpretability by highlighting the specific phenotypic features driving each diagnostic suggestion, which could be beneficial for clinician decision-making. Our findings indicate that integrating phenoBCBERT and PhenoGPT into clinical workflows holds promise for improving early detection and diagnosis of KS, potentially accelerating intervention and leading to better patient outcomes. This study underscores the potential of generative AI in rare disease diagnostics and paves the way for future applications in other complex genetic disorders Keywords - Phenotype Extraction ,Generative AI ,Natural Language Processing (NLP) ,PhenoBCBERT, PhenoGPT Transformer Models ,Clinical Decision Support Systems (CDSS) ,Electronic Health Records (EHRs),Human Phenotype Ontology (HPO) ,Data Augmentation