Critics: "Boo, -latest/greatest AI drug discovery method- only works if you have a similar training example!"
Biotech Industry: "Hurry, let's find the smallest possible modification to this approved ligand which hits the identical target and epitope"

Real data tracking my head nodding at the Autechre tour stop in Toronto last night

“What fascinated me was the fact that you could potentially predict which molecule could have this effect using thermodynamic principles in silica” www.fiercebiotech.com/special-repo...

"We discovered a TfR1-targeted brain shuttle, check out the brain uptake" (Crook et al. 2020, doi.org/10.1016/j.jm...). "This radioligand does not efficiently cross the blood-brain barrier" (Lin et al. 2025, doi.org/10.1007/s002...)

Love and respect the four-digit PDB code convention, and will miss it. I like how you can use it to roughly "date" structures. Like, 3J5P as a marker for the start of the cryo-EM membrane protein era in 2013.

Biology is hard example #103342: "Our in vivo screening efforts confirmed the requirement for high-affinity binding of the ligand for the cognate receptor to achieve robust activity" (*whisper* R^2 of 0.14) academic.oup.com/nar/article/...

Pharma: "Drug discovery is so hard, we have great binders but they fail in preclinical and clinical research"
AI Drug Discovery companies: "So... if we make better binders all diseases will be cured?"

Excellent study on peptide discovery and characterization, but I was not expecting the computational heavy-lifting of their prediction methodology to be... regex 😜 github.com/Svensson-Lab...

The art of writing a high impact manuscript, github.com/RosettaCommo...

In the VantAI webinar on Boltz-1 the authors describe challenges of arbitrary depositor-defined protein assemblies. Headaches incoming when they discover assembly is also dependent on experimental conditions 😅 E.g. www.biorxiv.org/content/10.1...

Classic general audience slide from David Baker (here, from his Nobel Prize talk). Would love to remake it to visualize the number of known folds for a 100-residue sequence, an estimate of possible folds (w/ sec. structure), and the number of disordered folds (no sec struct). Even more jaw-dropping?

Truth is stranger than fiction, in this case in the spirit of Willy Wonka. "Yes, on your left you'll see the sugar-coated peptide spaghetti lotus pond".

Every step forward in drug development is a victory in the battle against disease. Concept art for the boardroom of our future office. I just don't think we'll be able to secure lab space with an appropriate underground bunker...

Just a general PSA that getting a COVID booster before travelling is a great idea to avoid bringing back unwanted "biological souvenirs"...

There was nothing human about this particular green-over-chrome construct. Kassad noted the two horn-shaped ears, simple tunic, the profusion of thornspikes on the body, but not once did his gaze leave the two thousand-faceted eyes. "The Shrek", thought Kassad. "The Lord of Pain", whispered Moneta.

Love this, and ran it successfully on our GPU test node. However, I'd love to see some more examples. I was not able to modify the YAML file to introduce a modified residue. For example, this crashes with KeyError 'SME':

Pfizer med chem giving us reason #4392 why peptide design is hard. Even with a bound structure, constrained scaffold, and high activity hit; design is multi-objective, multi-state, and ensemble-based, in a large non-additive chemical space. "NMR-in-the-loop" required... pubs.acs.org/doi/abs/10.1...

People of sound mind: "It's not the SMS era, there's no reason to abide by the silly 140-character restriction for social media updates, be free!"
My bank: "Hold my beer"