Dr. Junseok Park is an Assistant Professor, computational scientist, and bioinformatician specializing in machine learning, large scale data pipelines, and multiomic analyses of patient cohorts. His current work focuses on developing clinical and genetics vision and language models and building pipelines to detect and prioritize pediatric rare disease variants by integrating structural variant genomic data with electronic medical records. His team develops classifiers to predict the pathogenicity of mobile element insertions and copy number variants, and fine tunes AI models on multimodal clinical data.
To advance these goals, the group is designing a structural variant image classifier trained using IGV style customized tracks to predict CNV pathogenicity, and optimizing scalable workflows on Google Cloud Platform, including AlphaGenome enabled resources, for high throughput structural variant annotation across regulatory features. In parallel, they are fine tuning large language models on multimodal electronic medical record data, including EEG, radiology, and clinical notes, to extract phenotypes and support genotype phenotype correlation, contributing to an end to end decision support platform for earlier rare disease diagnosis.