Blog posts

2026

Lab-in-the-Loop, and the Baselines It Has to Beat

2 minute read

Published:

Closed-loop, ML-guided biological design has produced real wet-lab results since 2019. It has also produced a corrections record, and every closed-loop result that has been independently re-examined has come back smaller. A review of what the campaigns measured, at what sample cost, and against which baseline.

AlphaGenome Atlas, and the Gap It Is Built On

5 minute read

Published:

DeepMind has precomputed the predicted effect of all 9 billion possible single-letter changes in the human genome. It is the third time they have run this play — and the first time on a problem where the models’ best-documented weakness is exactly the thing the table is for.

Zarr v2 vs v3: What Actually Changed, and When It Matters

7 minute read

Published:

Zarr v3 is not a cosmetic rewrite of v2. It reorganizes metadata, replaces the compressor with a typed codec pipeline, and adds sharding — which is the change that actually decides whether your dataset works on object storage. A practical comparison, and when each version is the right choice.

Graph Signal Processing: A Short History, and Who Is Building It

18 minute read

Published:

Where graph signal processing came from, how a field got named in 2013 by two papers that disagreed about what frequency means on a graph, how it reached biology, and the people whose work carries it now. With figures computed rather than drawn.

5 min knowledge from chatsheet: qptiff

less than 1 minute read

Published:

Visual cheat sheets on the TIFF family for microscopy — TIFF, OME-TIFF, QPTIFF, and how they compare. Click any thumbnail to zoom.