About

I am a computational biologist/bioinformatician. My primary research interests lie in using machine learning models to decipher how gene regulation is encoded in our genomes. I’ve also worked on variant effect and complex disease risk prediction.

Currently, I am a postdoc with Anshul Kundaje at Stanford, where I build model atlases to read and reprogram gene regulation. Previously, I did my PhD with Charles Danko at Cornell.

During my PhD, I developed CLIPNET, a deep learning model that predicts transcription initiation with single nucleotide resolution from DNA sequence. Interpretation of CLIPNET revealed a complex regulatory syntax governing how core promoter and transcriptional activator motifs work together to control when and where transcription begins. This work was published in Nature Genetics. At Stanford, I’ve scaled this approach up into an atlas of efficient, interpretable deep learning models trained on over 200 transcription initiation profiling experiments spanning all major human organs and many cancer types, which we use to map the sequence determinants of transcription initiation across diverse cellular contexts.

I’ve also worked on improving variant effect prediction by training deep learning models on personalized genomic sequences, and have contributed to studies using functional genomics to identify disease-driving variants in juvenile idiopathic arthritis.

I’m also passionate about open-source software, especially reviving and optimizing useful tools that have become unmaintained or difficult to run. For example, pydreg is a modern Python reimplementation of dREG that runs ~4.5x faster with ~5.4x lower peak memory (preprint); this work was supported by a Hidden Science Competition award from the Astera Institute. Similarly, LDetect-lite is a numerically equivalent and 60x faster re-implementation of LDetect, and gkmsvm-lite brings gapped k-mer SVMs to Python with GPU acceleration. You can find more of my software on my portfolio page.