Research Associate, Harvard T.H. Chan School of Public Health
Staff Scientist, Helmholtz Centre for Infection Research
My work focuses on machine learning methods for genomics, with an emphasis on model development, automated architecture search, and practical tools for microbial and genomic data.
Selected research outputs include Self-GenomeNet, GenomeNet-Architect, deepG, and work on expanding phenotype annotations for microbial species with language models.
Presents BAQLaVa for high-resolution viral profiling from metagenomic and metatranscriptomic data. The framework supports scalable virome epidemiology and virus-host interaction analysis.
Evaluates how well large language models support structured microbial phenotype annotation. The study compares model behavior across traits and uses confidence to prioritize reliable outputs.
Presents GenomeNet-Architect, a framework for automated neural architecture optimization on genomic sequence tasks. The approach searches model layouts and training hyperparameters for compact, efficient genomic classifiers.
Studies strain-level microbiome responses to repeated antibiotic perturbations in gnotobiotic mice. The paper highlights how treatment pulses can reshape community dynamics and resistance.
Introduces Self-GenomeNet, a self-supervised approach tailored to genomic sequences. The method uses reverse-complement structure to improve learning in data-scarce genomic prediction tasks.
Maps natural CRISPR-Cas systems across human microbiome metagenomes. The work links CRISPR spacers, cas genes, body sites, taxa, and putative viral or mobile-element targets.
For a complete list, see my Google Scholar profile.
I am the main developer of these tools, spanning secure research infrastructure, sequencing operations, and scientific writing support.
Context-aware AI support for scientific publishing, helping researchers revise manuscripts with cited references, domain language, and verifiable suggestions.
Policy-controlled sandboxes for coding agents on HPC systems, combining scoped filesystems, credential blocking, network policies, and auditable execution.
Open-source data capture and workflow coordination for sequencing facilities, supporting sample submissions, bioinformatics pipelines, and ENA publication.
The GenomeNet profile hosts pretrained models, datasets, and interactive Spaces for genomic and protein sequence analysis.
Interactive comparison of protein embeddings from ESM2 and Twin Network models for functional similarity analysis.
Browser-based extraction of DNA sequence embeddings from a pretrained metagenomic BERT model.
Interactive detection of CRISPR arrays in DNA sequences using a fine-tuned genomics model.
For collaborations, research questions, or speaking requests, email muench@hsph.harvard.edu or philipp.muench@helmholtz-hzi.de.
Angaben gemäß § 5 DDG
Dr. Philipp Münch - AI Software & SaaSE-Mail: mail@pmuench.com
USt-IdNr. gemäß § 27a UStG: DE460209447