De-Sheng Ker

Hard drives on backorder for two years as AI data centers trigger HDD shortage — delays forcing rapid transition to QLC SSDs The AI boom might help QLC overtake TLC in the next two years.

Better and faster microscope generates large amount of data which then lead to high resolution image. With AI server hunger for more data storage too, look like data storage will be next bottleneck in image processing.

Still this is pale in comparison with data generated by LHC per day.

This ‘landmark’ study describes the structural mechanism of strand exchange by the RAD51 filament using cryogenic structural, biochemical, and single-molecule analyses.
buff.ly/Ieh84gj

Structural mechanism of strand exchange by the RAD51 filament. Homologous Recombination (HR) preserves genomic stability by repairing double-strand DNA breaks and ensuring efficient DNA replication. Central to HR is the strand-exchange reaction taking place within the three-stranded synapsis wherein a RAD51 nucleoprotein filament binds to a donor DNA. Here we present the cryoEM structure of a displacement loop of human RAD51 that captures the synaptic state when the filament has become tightly bound to the donor DNA. The structure elucidates the mechanism of strand exchange by RAD51, including the filament engagement with the donor DNA, the strand invasion and pairing with the complementary sequence of the donor DNA, the capture of the non-complementary strand and the polarity of the strand-exchange reaction. Our findings provide fundamental mechanistic insights into the biochemical reaction of eukaryotic HR. ### Competing Interest Statement The authors have declared no competing interest.

Over five years of work has gone into our new preprint! The #cryoEM structure of a human RAD51 filament caught at the point of strand exchange 🧬
www.biorxiv.org/content/10.1...

TIL, New Zealand allows direct prescription drug advertising to consumer.

Join the first Keystone conference dedicated to DNA replication gaps and cancer! Send an abstract to be selected for a talk.

www.keystonesymposia.org/conferences/...

Just only discover the "--reuse_scratch" and "--keep_scratch" option in RELION, to minimize repetitively copying the same particle set into the cache directory for each job. I should have read more carefully at the long list of command options available.

Mastering the act to remotely wake up a workplace Linux machine from sleep using a smartphone at home with a single click.

By putting PC to sleep, is estimated to reduce the machine's power consumption by ten times when compared to its idle power consumption.

Graphormer/distributional_graphormer/protein at dig-v1.0 · microsoft/Graphormer Graphormer is a general-purpose deep learning backbone for molecular modeling. - microsoft/Graphormer

Source code available; easy to install; run test dataset without error; and job finish in less than a minute. 🤯

But it using a non-standard input and output format which require installation of additional software 😫

Spent another hour on installing it in GPU cluster with 80 GB VRAM A100, still getting CUDA out of memory..... Time for alternative approach

cryoSPARC benchmark of some GPU workstation in the department

mamba is faster than conda 🐍

Spent 1 hours trying to install a software, spent another 3 hours trying to work out how to use the software (poor documentation), only for the job to quit with an error: CUDA out of memory. 🥲

So it is possible to finish the RELION 2D classification with EM after VDAM using additional argument "--grad_em_iters 25"

Most cryoEM lab workstation: AMD ThreadRipper + 4090

Me: Ryzen 9 + 4070 Ti Super

Human Primosome (Initiation state)
doi.org/10.1111/febs...

#MolecularNode

Interesting,

(1xMPI, 64 Threads, --cpu) RELION's VDAM rans faster than (1xMPI, 32 Threads, 1 GPU, --gpu) RELION's VDAM in our HPC.

Also execute RELION's VRAM command with either mpirun or mpiexec are significant slower than srun command.

From Blue Letter F to Blue Birds, now to Blue Sky.