AlphaFold 2 isn’t just solving contact maps, it’s also mapping sequences → learned structure motifs. Clue: 2-3 seq-MSAs are sometimes enough🤔. Impacts:.- explains evo-contradictory fold-switch predictions, .- enables larger-scale ensemble mining, .- reveals a new AF input

RT @Align_Bio: 1/4.🚀 Announcing the 2025 Protein Engineering Tournament. This year’s challenge: design PETase enzymes, which degrade the ty….

A practical tip from the paper:. limiting sequence redesign to non‑interface residues preserves fold geometry and improves binding scores compared with full‑sequence optimization

Pretty engaging paper:. Fold-Conditioned De Novo Binder Design via AlphaFold2-Multimer Hallucination.

🏭FoldCraft turns AlphaFold2 into a binder factory: . a single contact‑map loss steers hallucination to any fold (Top7, β‑barrel, VHH. ) and beats RFdiffusion / RFAntibody with up to 20% in‑silico hit rates 🚀. 🏆 FoldCraft’s killer move: one contact‑map loss lets you dictate

Hm, protein sequence adds almost no extra predictive power once DNA & RNA models are fused, . while DNA+RNA synergy alone lifts performance by >10 % over any unimodal baseline (R² 0.584 vs 0.445). I guess not all modalities are equally valuable for every task and should guide

Apologies for the first line of this post if you love pipetting 😆. Read about it:. Rapid and Reproducible Multimodal Biological Foundation Model Development with AIDO.ModelGenerator.

Multimodal bio‑AI shouldn’t feel like pipetting code. AIDO.ModelGenerator lets you drag‑and‑drop 30 models (DNA 🧬, RNA 🧬, protein 🥩) via a single YAML. DNA+RNA fusion beats the isoform‑expression SOTA by >10 % and bumps SOX4 for Crohn’s up 6k ranks

RT @dr_alphalyrae: Really interesting experiment on how ‘AI scientists’ might interact with ‘flesh and blood scientists’ 👩🏻‍🔬 . Chatbots we….

Here's the ensemble sampling checklist I would consider after reading this paper:. 1. Prepare a single full-length MSA. 2. Run default (512:5120) + seven shallow depths. 3. Disable templates; enable multimer mode only if the protein is natively oligomeric. 4. Filter, cluster,.

Take-home: AF2 sometimes infers new folds by sequence association:. matching a query to structural patterns it learned during training, even when co‑evolution signals are absent. It's memory-driven, only needs shallow MSAs, and seems to be distinct from traditional homology

Read about it:. Large-scale predictions of alternative protein conformations by AlphaFold2-based sequence association.

The surprise here to me was that AFDistill regularization can double sequence diversity (up to +45 %) while improving structural fidelity. This challenges the usual diversity‑vs‑accuracy trade‑off in protein design

The paper:. AlphaFold distillation for inverse protein design.

AlphaFold is too slow for design loops at scale 🔄. AFDistill distills pTM/pLDDT into a sub-second, back-prop-ready feedback: +1-3 % recovery, up to +45 % diversity with no meaningful TM hit🏎️. I wonder whether this embedded “structure critic” trend holds🤔

While AF3’s accuracy spikes when accessory proteins inflate interface area, those contacts are functionally irrelevant for degrader design. once stripped, DockQ plummets for 90% of complexes. This exposes a scoring bias that can mislead model selection. I'm interested to see what

The paper:. Benchmarking the Builders: A Comparative Analysis of PRosettaC and AlphaFold3 for Predicting PROTAC Ternary Complexes.

🔥 PROTAC face‑off: AlphaFold3 vs PRosettaC. Benchmark ternary complexes against MD movies, not snapshots, and PRosettaC seems to beat AlphaFold3, despite 11/36 outright fails🤖💥🧪. Which tools do you think will lead the future of flexible, degrader‑aware AI?.

Interestingly, AlphaFold3’s chain‑pTM confidence scores often decouple from real accuracy for aptamers, especially short or novel folds . Several predictions with pTM < 0.3 still achieved sub‑2 Å RMSD, while some high‑pTM cases were wrong. This cautions users against relying

Read more:. Direct Modeling of DNA and RNA Aptamers with AlphaFold 3: A Promising Tool for Predicting Aptamer Structures and Aptamer–Target Interactions.