Research Scientist · Google Research
I work at the intersection of deep learning and genomics, developing methods that push the boundaries of variant calling, genome assembly, and clinical genomics. My research focuses on making genomic analysis more accurate, accessible, and impactful for precision medicine.
A draft human pangenome reference creates a new resource that better represents human genetic diversity. This allows scientists and doctors to more accurately study genetic variation across populations. This video covers the role of AI and deep learning — including DeepConsensus and DeepVariant — in enabling the construction of this groundbreaking resource.
Watch on YouTube →Selected invited talks and keynotes at leading conferences and institutions.
BICOB 2026 · Honolulu, Hawaii · March 2026
Keynote speaker at BICOB-2026, the premier international conference on bioinformatics and computational biology. Presenting recent advances in deep learning for genomic analysis.
MIT CSAIL Bioinformatics Seminar · Cambridge, MA · October 2025
Presenting DeepVariant, DeepSomatic, DeepPolisher, and pangenome-aware variant calling at the Theory of Computation group's Bioinformatics Seminar series.
ICLR 2025 · AI for Nucleic Acids Workshop · Singapore · April 2025
Invited talk at the International Conference on Learning Representations (ICLR), one of the top machine learning conferences, discussing deep learning methods for genomics including DeepVariant, DeepSomatic, and DeepConsensus.
Genome Institute of Singapore (A*STAR GIS) · April 2025
Seminar covering DeepVariant, DeepSomatic, DeepConsensus, DeepPolisher, and the opportunities presented by pangenome references to improve variant calling.
USC Dornsife · Los Angeles, CA · February 2025
Departmental seminar at the USC Dornsife Quantitative and Computational Biology program, presenting recent advances in deep learning for genomic analysis.
Selected publications in genomics, deep learning, and computational biology.
Nature Biotechnology, 2025
DeepSomatic is a deep-learning method for detecting somatic small nucleotide variations and insertions and deletions from both short-read and long-read data. Includes modes for whole-genome and whole-exome sequencing, tumor–normal, tumor-only, and FFPE samples.
Genome Research, 2025
An encoder-only transformer model for assembly polishing that predicts corrections to draft genome sequences using PacBio HiFi read alignments. Applied to 180 HPRC assemblies, achieving an average QV improvement of 3.4 (54% error reduction).
bioRxiv, 2025
A variant caller that integrates a pangenome reference alongside sample-specific read alignments within DeepVariant. Generates pileup images of both reads and pangenome haplotypes, reducing errors by up to 25.5% versus linear-reference-based DeepVariant.
Nature Communications, 2024
An approximate haplotagging method built directly within DeepVariant that locally haplotags long reads without requiring external phasing tools. Enables state-of-the-art variant calling across PacBio Revio, ONT R10.4 simplex, and duplex data.
Nature Methods, 2021
A haplotype-aware variant calling pipeline combining PEPPER, Margin, and DeepVariant that achieves high accuracy on nanopore long-read data across challenging genomic regions, advancing variant detection for both germline and clinical applications.