I’m Zong-Yan Liu (劉宗晏), a Ph.D. candidate in Plant Breeding and Genetics at Cornell University in the Buckler Lab (Cornell / USDA-ARS), with a minor in Computational Biology. I build deep learning and computational genomics tools that make plant genome interpretation more accurate, scalable, and reproducible.
I’m the lead author of GeneCAD, a sequence-only annotation system that turns raw genome sequence into complete, biologically coherent gene models — no RNA-seq, proteomics, or homology evidence required. I also contributed to the plant DNA foundation models PlantCaduceus (PNAS, 2025) and its long-context successor PlantCAD2 (Cell Genomics, 2026), pre-trained on 65 angiosperm genomes. In 2026 I serve on the ISMB program committee for Machine Learning in Computational and Systems Biology.
Beyond annotation, I’m taking these models downstream. reLoc uses protein language models to predict subcellular localization across plants, animals, and fungi — including when alternative isoforms of the same gene are sent to different compartments. And building on GeneCAD annotations of the 26 maize NAM founder genomes, I study genome evolution: dating the age of every maize gene and tracing how young and de novo genes arise.
What I work with:
- Deep learning — PyTorch, DNA language models (Caduceus / Mamba, ModernBERT), protein language models (ESM C, ProtT5), LoRA fine-tuning, CRF structured decoding
- Genomics — genome annotation (GFF3), comparative genomics & gene-age dating (phylostratigraphy, synteny), RNA-seq & small-RNA analysis
- Engineering — Python, R, GPU / HPC training pipelines, reproducible workflows, databases & web
