I recently reviewed a manuscript for a biotechnology journal I won’t name. The paper was clearly written by AI. I don’t mean edits or revisions here and there, I mean the entire manuscript. As if they had copied and pasted the entire contents of a certain research organization’s website into ChatGPT and asked for a paper on the topic.

I’m trying something new this week. I’ve been publishing weekly recaps on this newsletter for over a year now where I take a small deep dive into a few papers I’ve read recently. This takes time, and there’s often so much more interesting and relevant research and news in the data science + biotech space than I could possibly write about here. So I’ll be trying something different here for the next few weeks.

This week’s recap highlights analysis of human de novo mutation rates from a four-generation pedigree reference, how LLMs internalize scientific literature and citation practices, the py_ped_sim forward pedigree and genetic simulator for complex family pedigree analysis, and a review on predicting gene expression from DNA sequence using deep learning models like Enformer and Borzoi.

I recently wrote a piece about leaving academia for biotech. I left academia for industry in 2019. I spent four years at a consulting firm before joining Colossal Biosciences. This week I’m returning to the University of Virginia School of Data Science as a tenured associate professor and dean of research. The transition from academia to industry can be tricky, but it’s also increasingly common.

This week’s recap highlights nanoMDBG for metagenome assembly from nanopore reads, the SCassist AI-based workflow for single-cell analysis, discovery and characterization of GxE and GxG effects in a vertebrate model, the PIGEON framework for estimating gene-environment interaction for polygenic traits, and long-read alignment with multi-level parallelism.

I’ve written a lot about Ollama here. Ollama lets you run open-weight models like Llama, Gemma, Mistral, Qwen, DeepSeek, etc. on your own computer. You don’t have to pay for a frontier model like ChatGPT, Claude, or Gemini, and all the inputs and outputs stay on your computer, minimizing any privacy and security concerns. Until recently Ollama was a command-line only tool.

I liked Steve Krouse’s essay, “Vibe code is legacy code.” It helped crystalize some half-baked thoughts I have on vibe coding. Here’s an excerpt.Subscribe now Maintainability and vibe are inversely correlated I’ve been using GitHub copilot and chatbots for code for years, and I’ve written about them a lot here.

This week’s recap highlights Variant-EFFECTS for rewriting regulatory DNA to dissect and reprogram gene expression, zero-shot evaluation revealing the limitations of single-cell foundation models, EcoWeaver for large-scale prediction of gene functional associations from coevolutionary signals, and how assemblies of long-read metagenomes suffer from diverse errors.

I covered Autocycler (paper, code, docs) in last week’s recap: From the abstract: Here’s a schematic of the workflow: And some benchmarks:Subscribe now Demo I wanted to try this tool out myself. I followed the demo dataset described in the Autocycler docs, which contains ONT reads from a few E. coli plasmids, and mostly used the same code provided in the docs to run Autocycler on this data.

This week’s recap highlights Autocycler for long-read consensus assembly for bacterial genomes (future post on this one alone coming soon), Progen3 for broader generation and deeper functional understanding of proteins, the CarpeDeam de novo metagenome assembler for ancient datasets, and hifiasm ONT for efficient T2T assembly of Nanopore Simplex reads.

I started my academic career in biomedical research as faculty in the University of Virginia (UVA) School of Medicine. After eight years I jumped to industry &