About

I develop biologically informed AI models, AI agents, and GPU-accelerated algorithms to study how gene regulation shapes cellular states in cancer. My work integrates regulatory genomics and structure-based analysis to investigate transcription-factor interactions, interpret noncoding variants, and support biological discovery.

My PhD thesis introduced GET (General Expression Transformer), a foundation model achieving experimental-level accuracy in predicting gene expression across 213 human fetal and adult cell types. This work enables discovery of distal regulatory regions and transcription factor interactions linked to disease risk.

Research vision

I want to understand how interactions among transcription factors, DNA, and chromatin establish cellular states—and how these interactions can be altered in disease and therapeutics. Three questions guide my research.

How is regulatory cooperation encoded?

Red-chalk illustration of DNA wrapped around a nucleosome, with two interacting transcription factors.

How do protein interactions, the arrangement of DNA binding sites, and chromatin structure work together to control gene expression? Building on GET and our study on how TF-TF interaction can be encoded by dark genome repeats, I aim to connect models of transcription with physical mechanisms and understand how evolution distributes these regulatory arrangements across genomes.

Which regulatory changes drive disease?

Red-chalk illustration of a mutant protein with an altered interface that prevents binding to its partner.

Why does the same genetic change have different consequences across cell types? My work on noncoding risk variants, cancer mutations, and protein–DNA interactions motivates a broader goal: identify the regulatory interactions that sustain disease-associated cell states and determine which perturbations could change them.

Can AI make biological discovery faster and more systematic?

Red-chalk sketch of a branching exploration tree, with one path emphasized through successive experiments.

I am interested in models and research agents that connect predictions to biological mechanisms and help select informative experiments.

Selected work

Nucleosome-scale p53–hormone receptor grammar links cancer and primate evolution

Fu, X.*†, Cai, Q.*, Satya, P., Rabadan, R.†, and Levine, A. J.†

Unpublished manuscript (2026)

A foundation model of transcription across human cell types

Fu, X.*†, Mo, S.*, Buendia, A.*, et al.

Nature 637, 965–973 (2025)

Whole-genome analysis of noncoding genetic variations identifies multiscale regulatory element perturbations associated with Hirschsprung disease

Fu, A. X., Lui, K. N.-C., Tang, C. S.-M., et al.

Genome Research 30, 1618–1632 (2020)

*Equal contribution; †corresponding author.

All publications