About

My research spans thermodynamics, bioinformatics, and machine learning, with collaborations in geochemistry, microbiology, and database and software development.

Selected papers:

• BreCol: Benchmarking classical and deep-learning methods for microbiome-based cancer detection
  arXiv preprint, 2026

  Classical methods outperform current deep-learning approaches on holdout studies. A comparison of cancer types shows that colorectal cancer is easier to detect than breast cancer. Data and code are available in the BreCol repo.

• Chemical features of proteins in microbial genomes associated with body sites and gut inflammation
  Biomedical Informatics, 2025

  Oxidation and hydration state are the main chemical features that distinguish microbial genomes across body sites and inflammatory diseases. The methods are implemented in the canprot and chem16S packages.

• CHNOSZ: Thermodynamic calculations and diagrams for geochemistry
  Frontiers in Earth Science, 2019

  My flagship R package for thermodynamic calculations and diagrams, used across geochemistry, microbiology, and materials science. See data sources, demos, and documentation at the CHNOSZ website.

ML Evals and Apps

I bring an eval-driven mindset (“does it really work?”) together with deployment and usability.

That habit comes from research practices: maintaining R packages with automated tests and developing holdout benchmarks for bioinformatics. Thoughtful documentation and demos are what make my research software nice to use. I apply the same discipline to AI applications. Each one below has an evaluation set and runs as a live application.

• LangCalc

  A voice-driven calculator. A small language model is fine-tuned for multilingual function calling and scored on an evaluation dataset, and it runs on-device in an Android app.

• R-help-chat

  A retrieval-augmented generation system for an open-source software project. It answers questions from more than 10 years of R-help mailing list archives.

• AI4citations

  A citation verification system for scientific papers. It checks whether a cited source supports a claim.

Short Biography

Jeffrey Dick is a research scientist at Central South University, working on thermodynamic modeling, bioinformatics, and machine learning. He earned a PhD in Earth and Planetary Science from UC Berkeley, then held positions at Arizona State University and Curtin University before joining Central South University in 2017. During his PhD he created CHNOSZ, an open-source R package that has been maintained since 2009 and used by at least a hundred research groups worldwide. In 2025 he completed a machine learning engineering bootcamp at UC San Diego Extended Studies and began deploying ML applications on Hugging Face. He has also completed more than fifty peer reviews for academic journals and has taught in China and Thailand.