Research
Projects organized by biological scales, some projects span more than one scale.
Molecules
RNA–protein assemblies · Biomolecular structure · AI-predicted interfaces · Interactomes
Experimental characterization of viral proteins is costly and slow, and complexes predicted by generative AI still lack robust validation. As SARS-CoV-2 spread, my lead-author work built structural genomics and interactomics roadmaps of the virus from data on related coronaviruses, followed by a co-led commentary on improving models of protein complexes. PLI Analyzer (PLIA) now tests AI-predicted protein, RNA and DNA interfaces against known binding motifs and domains. Collaborative studies include the irCLIP-RNP and Re-CLIP methods for protein assemblies on RNA, and a review of this field.
- Bridging technical innovation and computational advances in studies of RNA–protein assemblies
- PLI Analyzer for data-driven validation of AI predicted biomolecular interfaces
- irCLIP-RNP and Re-CLIP reveal patterns of dynamic protein assemblies on RNA
- Computational protein modeling and the next viral pandemic
- Structural genomics of SARS-CoV-2 indicates evolutionary conserved functional regions of viral proteins
Cells
Single-cell transcriptomics · Cell-type discovery · Extrachromosomal DNA · Cancer gene regulation
Single-cell RNA-seq can resolve cell subtypes and states, but experimental noise and variation among cells of the same type challenge existing analysis methods. DUSC, my lead-author method, pairs deep-learning feature generation with model-based clustering to profile cell types robustly, and was applied to subclonal heterogeneity in triple-negative breast cancer. Contributions to collaborative cancer studies include work showing that extrachromosomal DNA generates and amplifies oncogene fusion transcripts, and that a mitochondrial pool of Raf1 drives glutamine catabolism to support tumour growth.
Tissues
Epidermis · Regulatory DNA variants · Transcription factors · Epithelial differentiation
Pinpointing which non-coding variants linked to polygenic skin disease disrupt gene regulation, and which transcription factors they affect, remains a central challenge. A collaborative study in human epidermis addressed this with massively parallel reporter assays and transcription-factor screens, and its data are shared through the ARVID-Skin portal, which I designed and developed. Related contributions examine how epidermal differentiation is controlled: glucose acting as a gradient signal that enables IRF6 dimerization, and opposing roles of the ubiquitin-like proteins NEDD8 and SUMO2 in homeostasis, regeneration and inflammation.
Organs & Organism
Trauma-related dissociation · Psychometrics · Female-biased autoimmunity · XIST ribonucleoprotein
Dissociation is under-diagnosed and associated with elevated suicidal self-injury, yet it has largely been left out of efforts to predict suicide. In women with trauma-related dissociation, my lead-author study used unsupervised learning to place patients along a dissociation spectrum and supervised models to predict prior suicide attempts. Autoimmune diseases also affect females disproportionately; contributions to collaborative studies showed that Xist ribonucleoprotein complexes promote female-biased autoimmunity, and that patients carry autoantibodies targeting hotspots on these complexes.
- Autoantibody hotspots reveal the origin and impact of immunogenic XIST ribonucleoprotein complexes in autoimmune diseases
- Xist ribonucleoproteins promote female sex-biased autoimmunity
- Unravelling psychiatric heterogeneity and predicting suicide attempts in women with trauma-related dissociation using artificial intelligence
Population
Outbreaks in confined settings · Agent-based modelling · Population-specific mutations · Inherited cancer risk
Outbreaks in confined settings such as cruise ships, schools and nursing homes spread rapidly through shared spaces and dense populations, and containment protocols can benefit from advanced computational models. My lead-author work developed the AI-GIS Infection Dynamics (AGID) agent-based model, which combines 3D maps of a confined environment with host behaviour and pathogen biology to predict daily infections and test containment protocols. In population genetics, collaborative studies profiled how population-specific mutations from the 1000 Genomes Project rewire the human interactome, and functionally tested inherited regulatory variants linked to cancer risk.