I study how new genes are born. Working across population genetics, single-cell and spatial-omics, biophysics, and evolutionary theory, I ask how previously inactive DNA evolves into functional genes de novo, and how genome organization, transcriptional noise, and molecular structure shape that process.
† corresponding author · ∗ equal contribution
bioRxiv (2026) preprint
bioRxiv (2026)
HP6/Umbrea is a fast-evolving Drosophila HP1 paralog that binds HP1a but cannot bind chromatin on its own. MD simulations show that HP6/Umbrea lowers dense-phase density inside HP1a condensates without shifting the phase-separation threshold, potentially operating as a "plasticizer" that may loosen heterochromatin and modulate silencing. Sequence analysis points to a recently evolved, constrained C-terminal tail underlying this role.
bioRxiv (2026) preprint
bioRxiv (2026)
Can brand-new (de novo) genes gain function without evolving complex binding partners? I show that thousands of segregating Drosophila de novo ORFs are enriched for bis-histidine (H-x-H) zinc-coordinating motifs, made when (CA) microsatellites translate into His-Thr-His repeats. Tracing ZMEG, a candidate zinc-binding ORF, from ancestral non-coding sequence, I argue that repeat expansion supplies a distributed class of candidate metal-binding peptides.
Trends in Genetics (2026)
Trends in Genetics (2026)
Machine learning, and deep learning in particular, is beginning to reshape evolutionary genetics and molecular evolution. This review surveys where ML now connects genotype, phenotype, and evolutionary history, from raw genetic variation to causal and multiomic approaches, and the challenges that remain.
Sustainable Microbiology (2026)
Sustainable Microbiology (2026)
Korean Natural Farming claims to cultivate and transfer beneficial "indigenous microorganisms" to soils, but the practice lacks validation. Using 16S and ITS sequencing across successive cultivation stages, we find the process is dominated by a few bacterial and fungal taxa and consistently reduces fungal diversity, contrary to its claimed benefit, though early cultures can capture and sustain bacterial diversity.
Proc. Natl. Acad. Sci. USA (2025)
Proc. Natl. Acad. Sci. USA (2025)
Comparing single-cell transcriptomes across three Drosophila species, we identify a core 198-gene set that robustly labels spermatogenic cell types across 25-30 My of divergence. Using it, we show how transcriptional bursting evolves through spermatogenesis, including conserved reductions in X-linked bursting, providing support for the cultivator model in which newly evolved testis transcripts locally tune neighboring genes' bursting.
Science Advances (2024)
Science Advances (2024)
Duplicate-gene models have largely ignored genome architecture. We show distally duplicated genes can be regulated by co-opting pre-existing 3D contacts ("enhancer capture"), using the young gene HP6/Umbrea as a case study. It captured a hidden enhancer (FLEE1) inside the essential gene MFS18, highlighting how enhancer capture is a one-step, evolvable solution to Ohno's dilemma.
Genome Biol. Evol. (2024)
Genome Biol. Evol. (2024)
How do genes arising de novo from non-coding DNA fix and gain function, especially in large populations where drift is weak? We propose a regulation-focused "cultivator" model in which each step in a de novo gene's trajectory is driven by selectable benefits to neighboring cultivator genes rather than the new gene itself, placing genome organization at the center of new-gene evolution.
BioSystems (2022)
BioSystems (2022)
We introduce a stochastic framework combining genetic and epigenetic contributions to phenotype to ask when plasticity evolves. Plasticity is favored transiently while adapting to new environments and erodes (canalizes) in stable ones, with selection choosing genic versus plastic routes by relative mutation rates, highlighting genetic conflict as a determinant of the long-term maintenance of phenotypic plasticity.
PLOS Genetics (2021)
PLOS Genetics (2021)
Are recently evolved genes dispensable? Knocking down 11,354 Drosophila genes, including 702 young ones (< 40 My), we find a similarly high proportion (~32%) essential among new and old alike, highlighting how new genes rapidly evolve essential developmental functions.
Science China Life Sciences (2019)
Science China Life Sciences (2019)
We analyze how integrating newly evolved genes reshapes human gene coexpression networks. Younger genes are more clustered and increasingly hierarchical, gaining connections through a "rich-gets-richer" process driven by duplication and orphan-gene origination. Despite continual new-gene integration, the network preserves its hierarchical, modular structure over evolutionary time.
eLife (2017)
eLife (2017)
How do circadian clocks lock onto daily cycles whose length changes with the seasons? Driving the cyanobacterial clock in vivo and in vitro, we find its phase follows a simple scaling law that tracks midday, intrinsic to the minimal KaiABC protein system. This framework based on cue-driven phase shifts can thus predict clock behavior across many environments.
Phys. Rev. E (2016)
Phys. Rev. E (2016)
Using an exactly solvable driven-dissipative model, we study nonequilibrium dynamics via information length. We find geodesics along which information moves at constant speed, giving optimal paths that minimize time and dissipated energy, and show small periodic modulations can control a stochastic growth model.
US Patent Application (2016) patent
US Patent Application (2016)
Methods for determining breast-cancer prognosis from a defined set of genes, computing a prognosis score via a specific algorithm. Also covers related compositions, kits, and treatment approaches for the most aggressive breast cancers.
PLOS ONE (2015)
PLOS ONE (2015)
Genetically identical clonal cells can still diverge phenotypically, with consequences from bet-hedging to drug resistance. Using stochastic models with finite correlation time, we show how noisy growth and loss of self-regulation shift populations from bounded to unbounded growth, and how variance, not just the mean, shapes phenotype.
PLOS ONE (2013)
PLOS ONE (2013)
Triple-negative breast cancers are aggressive, lack targeted therapies, and are hard to stratify. We built a 30-gene BACH1 Pathway Metastasis Signature (BPMS) that selectively predicts metastasis-free survival in basal-like/TNBC patients and refines risk beyond existing clinical tests. It works as a single-sample predictor and nominates BACH1-pathway genes as therapeutic targets.