I’m Ken Osumi (大角 健), founder of FoldArc.
What I write here
I read the internal representations of protein language models — ESMC and its sparse-autoencoder features — and try to turn them into residue-level, falsifiable hypotheses about molecular mechanism. This blog is where I do that in the open: reproductions, methods, and the occasional wrong guess. The bet behind it is simple — a model trained only on sequence has quietly learned a great deal of biology, and most of it is still waiting to be read.
Background
My training is in structural and chromatin biology. I did my PhD at the University of Tokyo in Hiroshi Kurumizaka’s lab, using cryo-EM to study how RNA polymerase II transcribes through nucleosomes and recognizes damaged nucleotides; before that, in the Siomi lab, I worked on transposon silencing and the regulation of SETDB1. I was a JSPS Research Fellow (DC2) and a University of Tokyo WINGS-LST fellow.
Selected publications
- Osumi K, Kujirai T, Ehara H, et al. Structural Basis of Damaged Nucleotide Recognition by Transcribing RNA Polymerase II in the Nucleosome. J Mol Biol 435(13):168130 (2023).
- Osumi K, Sato K, Murano K, Siomi H, Siomi MC. Essential roles of Windei and nuclear monoubiquitination of Eggless/SETDB1 in transposon silencing. EMBO Reports 20(12):e48296 (2019).
- Akatsu M, Ehara H, Kujirai T, et al. (incl. Osumi K). Cryo-EM structures of RNA polymerase II–nucleosome complexes rewrapping transcribed DNA. J Biol Chem 299(12):105477 (2023).
FoldArc
The day job. FoldArc designs de novo protein binders — small proteins you can express inside a cell to switch off a single protein–protein interaction while leaving everything else intact. De novo protein probes for cell biology. This blog is the other half of that work.