I'm building AI systems to protect against biological threats.
At Valthos, I lead research and product development to identify biological threats and design medical countermeasures in real time. I live in New York City.
Previously, I designed RNA-based medicines at Inceptive using AI and high-throughput experiments. As a group leader at the Broad Institute and a graduate researcher in the Sabeti Lab, I developed genomic tools to detect pathogens and track their evolution at scale. At MIT, I earned an SB in computer science and in physics, followed by an MEng and PhD in computer science.
Viral diagnostics that are highly sensitive across genomic variation.
ADAPT rapidly designs sensitive and specific CRISPR-Cas13–based diagnostic assays for any virus, factoring in publicly available genome sequences. It powers a resource of diagnostic assay designs for more than 1,900 vertebrate-infecting viral species, and ADAPT-designed assays outperformed existing design methods in experimental testing.
Target-enrichment panels that enable sensitive metagenomic sequencing across broad genomic diversity.
CATCH designs panels for enriching diverse genomic targets prior to sequencing, making metagenomic sequencing more sensitive and improving genome recovery without prior hypotheses about sample contents. It is widely used for pathogen surveillance and other applications involving diverse targets, such as HLA typing.
My full publication list is on Google Scholar. Below are a few selected papers from projects I led.
This project developed methods to generate artificial CRISPR-Cas13a guide RNAs, including guides deliberately mismatched to their targets. In experimental testing, they improved detection across target sequence diversity and discrimination between closely related variants.
This work was the first application of machine learning to viral diagnostic design. It involved creating a training dataset on diagnostic sensitivity, learning a model of sensitivity, and developing design methods that account for viral genomic diversity. In experimental testing, the resulting assays outperformed existing design paradigms.
Coverage: Broad Institute news
This work introduced an algorithm for designing panels that enrich extensive genomic diversity, such as hundreds of viral species and all their known variation, prior to sequencing. In experimental testing, a CATCH-designed panel targeting all 356 viral species then known to infect humans increased viral content 18-fold and recovered genomes that could not otherwise be assembled.
Coverage: Broad Institute news
This study generated the largest and most diverse Zika genome dataset of the 2015–16 epidemic, mapped the virus's spread across the Americas, and showed that Zika had circulated undetected for months, highlighting the need for viral surveillance.
Coverage: Broad Institute news Nature News & Views Time The Atlantic Wired NBC News