Date: 10 December 2025 (Wednesday)
Time: 11:00 am – 12:00 pm
Venue: Seminar Room 1, G/F, Laboratory Block, 21 Sassoon Road
Host: Professor Yuanhua Huang

Dr. Qiao Chen is a Senior Deep Learning Engineer at Illumina’s Artificial Intelligence Lab, where he develops AI-driven methods for genomic interpretation, scientific literature curation, and clinical evidence mining. He is interested in answering research questions in various domains across social behavioral and natural sciences, from a computational perspective. Before joining Illumina, Chen completed a postdoctoral fellowship at the University of Hong Kong’s School of Biomedical Sciences, focusing on computational approaches for single-cell and spatial transcriptomics, including RNA-velocity modeling and spatial-pattern imputation. He holds a PhD from HKU, where he developed probabilistic and neural models for learning analytics and automatic text assessment.

Modern biomedical data analysis benefits from effective representation learning, illustrated by three initiatives in this talk. First, we introduce SeededNTM, a neural topic model for learning topic representations for spatial transcriptomics, which uses marker gene–driven topic seeding to guide cell type deconvolution without external reference data. Second, we develop a semi-supervised strategy for phenotyping UK Biobank imaging data, with contrastive learning on Liver MRIs to derive image‐derived phenotypes that enable novel genetic discoveries beyond classical traits. Third, we learn phenotype-aware gene embeddings from gene-gene interaction networks and show promising performance in the downstream task of prioritizing diagnostic variants for rare diseases. These approaches demonstrate how learned representations improve interpretability and discovery across scales of biomedical data, from population to omics scales.