What I work on
I develop machine learning methods for genomics — from DNA language models that predict gene expression to algorithms for genome annotation, assembly, and indexing.
Gene expression
Shorkie is a masked-DNA language model pretrained on 165 fungal genomes and fine-tuned on high-resolution yeast transcription-factor induction time-course RNA-seq data to predict dynamic expression patterns.
RNA splicing
A dilated residual convolutional neural network that decodes RNA splicing, alternative splicing, and the impact of genetic variants on cryptic splicing — released as Splam and OpenSpliceAI.
Genome annotation
Graph-based methods that combine fragmented DNA and protein alignments to assemble more accurate genome annotations, implemented in the LiftOn annotation lift-over tool.
Genome assembly
Han1: the first gapless, reference-quality, fully annotated Southern Han Chinese genome, assembled from PacBio HiFi and Oxford Nanopore long reads using T2T-CHM13 as a guide.