Lab Newsletter — July 29, 2026: Seeing the Cell's Machines

Most of this newsletter lives in the world of sequences — genomes, transcripts, proteins as strings. Today we go the other way, all the way down to the picture: cryo-electron tomography (cryo-ET) can now photograph molecular machines in their native cell, unpurified and in place. The long-standing dream is visual proteomics — a molecular atlas of the cell read directly from these images. The catch was never the microscope. It’s that a single tomogram is a crowded, near-noise 3D volume where most macromolecules “are hardly discernible from noise,” and someone has to find and name every one. This week, that someone is increasingly an AI — trained on open, community-built data.

🔬 An open atlas you can download

The anchor is a striking act of open science. A Molecular Cell study (Plitzko, Engel & Kotecha labs) released 1,829 annotated in-cell tomograms of the green alga Chlamydomonas reinhardtii — prepared by cryo-plasma-FIB milling, spanning the cell’s organelles, raw data and all — explicitly as a community resource. To prove the dataset’s worth they averaged complexes across a staggering size range, from >3 MDa ribosomes down to ~200 kDa (Rubisco, nucleosomes, clathrin, photosystem II, ATP synthase), with most maps reaching sub-nanometer resolution. The raw tilt-series sit in EMPIAR and the annotations on GitHub. Why it matters for the lab: this is imaging-as-measurement at the ultimate resolution — the structural ground truth beneath whole-cell modeling — and it’s shared the way we believe data should be: large, annotated, and open for anyone to build on.

🤖 Crowdsourcing the bottleneck

If picking particles by hand “can take months,” why not turn it into a sport? That’s what the Chan Zuckerberg Imaging Institute did: a 3-month Kaggle challenge to annotate six molecular species across hundreds of experimental tomograms drew over 1,000 participants and produced particle pickers that beat the previous state of the art (a Nov 2025 write-up found data augmentation was the decisive trick). Everything — tomograms, ground truth, and the winners’ models — was released CC0 on the CryoET Data Portal, alongside open tools like Copick and MONAI 3D-U-Net notebooks, and the dataset now lives on CZI’s Virtual Cells Platform. Why it matters for the lab: a crowdsourced, openly-benchmarked model that outruns the specialists is a BioImage Model Zoo story in a new modality — and it’s no accident the data landed on a virtual-cell platform. Visual proteomics is one of the atlases a virtual cell will be built from.

🧩 AI along the whole pipeline — and the honest frontier

Zoom out and the whole cryo-ET workflow is quietly becoming a deep-learning stack. A July 2025 review (with UCLA’s Z. Hong Zhou and vision pioneer Demetri Terzopoulos) maps AI onto every stage: picking (Topaz, crYOLO, CryoSegNet), denoising and missing-wedge repair (Topaz-Denoise, IsoNet), orientation-bias correction (spIsoNet, cryoPROS), and automated model building (ModelAngelo, DeepTracer, CryoREAD) — turning intractable, noisy volumes into interpretable structures “from HIV virus-like particles to in situ ribosomal complexes.” Open toolkits like AITom (Min Xu’s CMU group) package the in-cell half end-to-end. The honest frontier: only a handful of molecular species can yet be reliably identified in the crowded, low-contrast cytoplasm — the full “molecular atlas of the cell” is still a horizon, not a result. Why it matters for the lab: this is our thesis in a different key — open images, open models, community benchmarks — the same stack we build for light microscopy (BioEngine, AI4Life), now pointed at molecules in situ.

Sequences tell you what a cell could build; visual proteomics shows you what it actually did, and where. Getting there is a computer-vision problem as much as a microscopy one — which is exactly why it’s on-brand for a lab that lives where imaging and AI meet.

Sources linked inline. Compiled by Happy Agent; the lab footer notes our AI-assisted content. (X/Twitter sweep was skipped today — our news API is out of credits.) Have lab news to share — a talk, paper, conference or release? Message me on Slack.