Conformational ensemble-based framework enables rapid development of Lassa virus vaccine candidates biorxiv.org/cgi/content/sh… #biorxiv_immuno

Conformational Ensemble-Based Framework Enables Rapid Development of Lassa Virus Vaccine Candidates • This study introduces an AI-driven framework combining subsampled AlphaFold2 and ProteinMPNN to stabilize the conformationally dynamic Lassa virus glycoprotein complex (GPC), a

The concept of critical batch size is quite simple. Let’s assume we have a training dataset with 1M tokens. If we use a batch size of 10, we can update model param 100,000 times. On the other hand, if we increase the batch size to 100, the step size decreases to 10,000 (1/n).

More fantastic work by the amazing Sarah Burbach

ABCFold: Easier Running and Comparison of AlphaFold 3, Boltz-1, and Chai-1 - Structural biology has seen a revolution with deep learning-based protein structure predictors like AlphaFold 3, Boltz-1, and Chai-1. However, running and comparing these models efficiently remains a

Thrilled that our work on this problem with Karel Břinda, Zamin Iqbal, and others is out in Nature Methods today! We used phylogenetic compression (described in the thread) to compress every microbe ever sequenced onto a flash drive so that it can be searched with a laptop!

Since BoltzDesign1 just got published, people might also be interested in github.com/escalante-bio/… by Nick Boyd It has a nicely flexible system for including custom losses. It's also surprisingly little code on top of Boltz (and the Jaxified Joltz).

Our several-years-old fix to ProteinMPNN's tendency to make weird antibody CDR seqs is finally out. We run an antibody LM in parallel & added its logits to ProteinMPNN's, fixing most issues we encountered. It also increased % of HER2-binding trastuzumab designs >10-fold

Better antibodies engineered with a GLIMPSE of human data １．GLIMPSE-1 is a new antibody language model trained solely on paired human antibody sequences. It achieves state-of-the-art results in humanization tasks, rivaling models trained on far larger and more diverse

A fantastic collaborative effort with the BatistaLab (Ragon/Harvard), ShawLab(Penn) and IrvineLab (Scripps) where we show how targeted immunogen design can induce HIV broadly neutralizing antibody responses! biorxiv.org/content/10.110…

i am once again asking Weights & Biases to make a phone app so that i can monitor the situation when i'm outside

considering Muon is so popular and validated at scale, we've just decided to welcome a PR for it in PyTorch core by default. If anyone wants to take a crack at it... github.com/pytorch/pytorc…

Come on, is it that hard to order wet lab experiments?

Mixture of Raytraced Experts Stacked MoE architecture that can dynamically select sequences of experts, producing computational graphs of variable width and depth. Allows predictions with increasing accuracy as the computation cycles through the experts' sequence. Links below

"Enhancing sequence alignment of adaptive immune receptors through multi-task deep learning" by Gur Yaari lab. Very promising: speed boost from GPU acceleration + solving problems of greedy aligners in capturing V(D)J rearrangement uncertainty academic.oup.com/nar/article/53…

🚀Protenix‑Mini: trims 32 / 48 early Pairformers and swaps AF‑style 200‑step diffusion for a 2‑step ODE (η = 1, γ₀ = 0)👀 This cuts ≈85% inference FLOPs & 70% memory, with only ~2% LDDT hit on RecPDB🔥

CoV-UniBind: A Unified Antibody Binding Database for SARS-CoV-2 biorxiv.org/content/10.110… #biorxiv_immuno

We’re open-sourcing the pre-training code for Phi4-mini-Flash, our SoTA hybrid model that delivers 10× faster reasoning than Transformers — along with μP++, a suite of simple yet powerful scaling laws for stable large-scale training. 🔗 github.com/microsoft/Arch… (1/4)

In 1965, Margaret Dayhoff published the Atlas of Protein Sequence and Structure, which collated the 65 proteins whose amino acid sequences were then known. Inspired by that Atlas, today we are releasing the Dayhoff Atlas of protein sequence data and protein language models.

For protein science/bioinformatics: MMseqs Diamond FoldSeek BioTite ColabFold BioPython pdb-tools Seqkit hh-suite IQ-Tree MAFFT TMAlign PyRosetta (Just released in 2024) DSSP BLAST ProteinMPNN RFDiff AlphaFold2 ESM BindCraft RDKit Boltz HMMER GROMACS Which others?