Om (@decodediscovery) header
Om (@decodediscovery) profile picture

we help you launch discovery programs with a query.

La Jolla
Joined January 2024
1/ 🧪 a de novo enzyme that oxidizes one chosen C-H bond on a steroid: @UWproteindesign, @OfficialUoM and collaborators designed heme hydroxylases that hit estradiol at either the C6 (93% d.e.) or C16 (99% d.e.) position. 👇
1
10
1/ 🧬 can a protein folding model tell a real antibody binder from a lookalike decoy? a new @Roche Prescient Design benchmark says better than chance, far from reliable. it ships as an open dataset built from patents. 👇
1
1
10
1/ 🧬 a new structure model reports 79.07% on FoldBench antibody-antigen interfaces vs 47.90% for AlphaFold 3: NovaAtom-Lite from @Geodesiclab, benchmarked by the company. 👇
2
21
1/ 🧪 9 of 19 AI-designed peptides bound their target in the lab, and 2 beat the natural ligand's apparent affinity. @try_litefold's LiteMol-1 designed them from the BHRF1 protein sequence alone. 👇
1
1
8
“We will have them fight against each other and decide what’s real, what’s not, review all the data.” @ealarkin7, founder and CEO of @Valius_Sciences, on how his team has several AI models argue over each cancer patient’s tumor data and weigh the evidence. 🎙️
1
16
1/ 🧪 112 of 210 designed protein binders were active in a lab binding assay, across 10 targets. Levinthal's Pallatom2, a 65.9M-parameter model that designs sequence, backbone, side chains and target contacts together. 👇
1
8
1/ 🧬 which co-folding model predicts antibody-antigen complexes best? a new @imperialcollege benchmark in mAbs ran five on 200 complexes: OpenDDE had the top median DockQ, and it and Protenix-v2 both beat AlphaFold 3. 👇
1
8
1/ 🧬 allosteric-site predictors trained on ligand-bound structures miss pockets that only show up in other states. a new bioRxiv model, AllositeUltra, finds sites across apo and substrate-bound forms, then validated a new one in the wet lab. 👇
1
15
1/ 🧪 a model that reads a standard H&E pathology slide and predicts a single-cell spatial map of protein expression. Hist2Prot is out in @NatureCancer, with public code. 👇
1
18
Replying to @davidrliu
@davidrliu, the Broad Institute chemist who invented base and prime editing, on how deeply AI now runs through gene editing research: "our lab and CTG uses AI for just about every aspect of the whole enterprise as a tool."
1
27
1/ 🧬 antibodies usually fail when their target mutates. @Stanford and @arcinstitute built Germinal Multi to design one antibody against many variants of a target at once, including variants held out of the design. 👇
1
14
1/ 🧪 a model that flags siRNA off-target hits before the wet lab: @DeepGenomics' DeepRNAi predicts on- and off-target knockdown straight from siRNA and mRNA sequence, and they released a public benchmark with it. 👇
1
17
1/ 🧪 asked which of two predicted binding poses was right, expert chemists did no better than chance. an open-weight 27B model, after RL training, reached 74%. that's Pyrite, a new benchmark from @genesismolai and Stanford. 👇
1
14