Why we built Shorkie: reading the regulatory code of yeast
Research summary9 min read
The story behind Shorkie — a fungal DNA language model that learns the regulatory grammar of budding yeast and predicts how its genes turn on and off over time.
Writing
Posts
Research summaries and opinions — notes on the work I do and the ideas behind it.
Why we rebuilt SpliceAI: an open, retrainable splice-site predictor
Research summary7 min read
The story behind OpenSpliceAI — a faithful, fast PyTorch reimplementation of SpliceAI that anyone can retrain on any species, and use to read the splicing impact of genetic variants.
Why we built LiftOn: better genome annotation by combining DNA and protein
Research summary10 min read
The story behind LiftOn — a genome-annotation tool that fuses DNA alignment (Liftoff) and protein alignment (miniprot) so you can carry accurate gene annotations onto a new assembly, even across species.
Why we built Splam: a splice-junction model that cleans up RNA-seq
Research summary6 min read
The story behind Splam — a compact deep-learning model that scores a splice junction from its local sequence, more accurately than SpliceAI, and uses it to strip spurious alignments out of RNA-seq.
Why we built WGT: tools for Wheeler graphs, the backbone of pangenome indexes
Research summary8 min read
The story behind the Wheeler Graph Toolkit — software to generate, recognize, and visualize Wheeler graphs, the elegant class of graphs that can be indexed and searched like a BWT, using a fast heuristic and an SMT solver.
Why we built Han1: a complete, annotated genome for the world's largest ethnic group
Research summary6 min read
The story behind Han1 — the first gapless, reference-quality, fully annotated human genome from a Han Chinese individual, and what comparing two finished human genomes revealed.
Why we built sangeranalyseR: painless Sanger sequencing analysis in R
Research summary6 min read
The story behind sangeranalyseR — a free, open-source R/Bioconductor package that takes you from raw .ab1 chromatograms to aligned consensus sequences in about four lines of code, with an interactive Shiny GUI.
No posts match your search.