Joe Greener
@jgreener64.bsky.social
Computational chemist/structural bioinformatician working on improving molecular simulation at MRC Laboratory of Molecular Biology. jgreener64.github.io
Round 1 of our machine learning potential really wants oxygen atoms in water to cluster together 😂
May 14, 2025 at 12:42 PM
Round 1 of our machine learning potential really wants oxygen atoms in water to cluster together 😂
Had a great time at the Museu de les Ciències Príncipe Felipe in Valencia, Spain.
Saw lots of kids engaged with science and the Chromosome Forest exhibition covering much of biology. I can just about forgive the artistic licence in the DNA model.
Saw lots of kids engaged with science and the Chromosome Forest exhibition covering much of biology. I can just about forgive the artistic licence in the DNA model.
April 20, 2025 at 3:44 PM
Had a great time at the Museu de les Ciències Príncipe Felipe in Valencia, Spain.
Saw lots of kids engaged with science and the Chromosome Forest exhibition covering much of biology. I can just about forgive the artistic licence in the DNA model.
Saw lots of kids engaged with science and the Chromosome Forest exhibition covering much of biology. I can just about forgive the artistic licence in the DNA model.
When you're training a force field and step 2 of the simulation looks like this.
March 31, 2025 at 5:41 PM
When you're training a force field and step 2 of the simulation looks like this.
One surprise is how similar the gradients are between reversible simulation and ensemble reweighting, two different approaches.
We do see evidence that reversible simulation gives more stable gradients over repeats.
(4/8)
We do see evidence that reversible simulation gives more stable gradients over repeats.
(4/8)
December 6, 2024 at 12:16 PM
One surprise is how similar the gradients are between reversible simulation and ensemble reweighting, two different approaches.
We do see evidence that reversible simulation gives more stable gradients over repeats.
(4/8)
We do see evidence that reversible simulation gives more stable gradients over repeats.
(4/8)
Differentiable molecular simulation is promising for training force fields but has 3 problems: high memory cost, slow run times and unstable gradients.
Here we solve these by explicitly deriving and computing the gradients from AD.
(2/8)
Here we solve these by explicitly deriving and computing the gradients from AD.
(2/8)
December 6, 2024 at 12:16 PM
Differentiable molecular simulation is promising for training force fields but has 3 problems: high memory cost, slow run times and unstable gradients.
Here we solve these by explicitly deriving and computing the gradients from AD.
(2/8)
Here we solve these by explicitly deriving and computing the gradients from AD.
(2/8)