澪乃ゆい@Bioinfo VTuber

FrustrAI-Seq: Scaling Local Energetic Frustration to the Protein Sequence Space Jan-Philipp Leusch, Miriam Poley-Gil, Miguel Fernandez-Martin, Nicola Bordin, Burkhard Rost, R. Gonzalo Parra, Michael Heinzinger bioRxiv 2026.02.03.703498; doi: https://doi.org/10.64898/2026.02.03.703498 from Figure 1 (CC BY 4.0)

FrustrAI-Seq predicts local energetic frustration from sequence alone. ProtT5 with LoRA fine-tuning trained on ~1M proteins analyzes the human proteome in 17 min, extracting physical energy constraints embedded in evolutionary statistics and enabling de novo protein design.
doi.org/10.64898/202...

06.02.2026 23:21 — 👍 1    🔁 1    💬 0    📌 0

Protenix-v1 white paper

github.com/bytedance/Pr...

04.02.2026 22:02 — 👍 0    🔁 0    💬 0    📌 0

ByteDance developed Protenix-v1, achieving AlphaFold3-level structure prediction. Accuracy improves via inference-time scaling, with RNA MSA and template support. It reaches 52.3% success on FoldBench (Ab–Ag) and 79.4% on PXM-2024, and highlights evaluation issues while providing a new benchmark.

04.02.2026 22:02 — 👍 0    🔁 0    💬 1    📌 0

GitHub - cddlab/boltz_restr: Boltz-1/2 with restraint-guided inference (Ligand Conformer Restraint and Distance Restraint) Boltz-1/2 with restraint-guided inference (Ligand Conformer Restraint and Distance Restraint) - cddlab/boltz_restr

boltz_restr extends Boltz-1/2 inference with distance restraints. It reuses pretrained weights without retraining, adds ligand conformer and distance constraints, and enables GPU-accelerated sampling of protein–ligand complexes and dissociation.

github.com/cddlab/boltz...

04.02.2026 21:41 — 👍 0    🔁 0    💬 0    📌 0

STAR-MD: Long-Horizon Protein Dynamics Generation First model to demonstrate stable, physically plausible generation out to the 1 microsecond timescale.

bytedance-seed.github.io/ConfRover/st...

04.02.2026 14:00 — 👍 0    🔁 0    💬 0    📌 0

ConfRover: Generate Protein Conformational Dynamics A unified model for protein ensembles and dynamics.

ByteDance SEED released generative models for protein dynamics. ConfRover learns MD trajectories autoregressively, while STAR-MD uses spatiotemporal attention for microsecond-scale generation. STAR-MD improves structural validity by 64% and RMSD by 65%, reproducing realistic dynamics.

04.02.2026 14:00 — 👍 0    🔁 0    💬 1    📌 0

ArtiDock: Accurate Machine Learning Approach to Protein–Ligand Docking Optimized for High-Throughput Virtual Screening Classical protein–ligand docking has been a cornerstone technique in computational drug discovery for decades but has reached an accuracy and performance plateau. Recently introduced Machine Learning (ML)-based docking methods offer a promising paradigm shift, but their practical adoption is hampered by accuracy-to-speed trade-offs, inadequate benchmarking standards, and questionable chemical validity of predicted poses. In this study, we introduce ArtiDock─an ML-based docking technique optimized for high-throughput virtual screening applications. To evaluate ArtiDock, we developed a dedicated performance and accuracy benchmark for pocket-specific rigid protein–ligand docking, which mimics realistic industrial drug discovery scenarios and is based on the novel PLINDER data set. We demonstrate that ArtiDock is 29–38% more accurate in comparison to leading open-source and commercial classical docking techniques such as AutoDock, Vina, and Glide, while providing a low computational cost. ArtiDock notably excels in challenging docking scenarios involving unbound protein structures and binding sites containing ions and structured water molecules. Additionally, we demonstrated competitive accuracy of our approach at considerably higher throughput compared to a wide range of AI docking and AI cofolding methods using the PoseX benchmark. Our results show that ArtiDock could be considered as a method of choice in high-throughput virtual screening scenarios.

ArtiDock, a protein–ligand docking tool, achieves 29–38% higher accuracy on the PLINDER benchmark. Docking 1M compounds costs about $5. It performs well on apo structures and ion/water-containing pockets, making it practical for large-scale screening.

pubs.acs.org/doi/10.1021/...

04.02.2026 13:56 — 👍 0    🔁 0    💬 0    📌 0

I am a Bioinfo VTuber who explains bioinformatics topics in Japanese,
with a focus on protein structure prediction
and protein design models.

This Bluesky account is used to share selected insights
and connect with an international audience.

www.youtube.com/@mionoyui

04.02.2026 00:47 — 👍 2    🔁 0    💬 0    📌 0

Toward Interpretable and Generalizable AI in Regulatory Genomics Deciphering how DNA sequence encodes gene regulation remains a central challenge in biology. Advances in machine learning and functional genomics have enabled sequence-to-function (seq2func) models th...

This is a review of seq2func models. It analyzes why high accuracy on held-out data does not guarantee generalization under perturbations, and proposes a causal refinement framework that combines active learning with targeted perturbation experiments and continual learning.

arxiv.org/abs/2602.01230

04.02.2026 00:29 — 👍 0    🔁 0    💬 0    📌 0

Decoding Protein–Membrane Binding Interfaces from Surface-Fingerprint-Based Geometric Deep Learning and Molecular Dynamics Simulations Predicting protein–membrane interactions is a formidable challenge due to the subtle physicochemical features that distinguish membrane-binding regions of a protein surface as well as the scarcity of experimentally resolved membrane-bound protein conformations. Here, we present MaSIF-PMP, a geometric deep learning model that leverages molecular surface fingerprints to predict interfacial binding sites (IBSs) of peripheral membrane proteins (PMPs). MaSIF-PMP integrates geometric and chemical surface features to produce spatially resolved IBS predictions. Compared to existing models, MaSIF-PMP achieves superior performance for IBS classification, while feature ablation studies and transfer learning analyses reveal distinct determinants governing protein–membrane versus protein–protein interactions. We further show that molecular dynamics (MD) simulations can validate model predictions, refine IBS labels, and capture composition-dependent membrane binding patterns. These results establish MaSIF-PMP as an effective framework for IBS prediction and highlight the potential of incorporating conformational dynamics from MD to improve both the model accuracy and biological interpretability.

MaSIF-PMP, a geometric deep learning model for protein–membrane interface prediction, learns 5 surface features and achieves ROC AUC 0.78, outperforming existing methods. Integration with HMMM simulations enables membrane composition analysis.

pubs.acs.org/doi/10.1021/...

04.02.2026 00:27 — 👍 0    🔁 0    💬 0    📌 0

From Serendipity to Strategy: Rationalizing Molecular Glue Discovery and Proximity-Induced Pharmacology through Chemical Biology Molecular glues represent a new paradigm in drug discovery, stabilizing novel protein–protein interactions between two proteins to elicit targeted cellular outcomes. Historically discovered through se...

Review on molecular glues
Chemical space exploration (bias libraries, DEL, covalent probes), PIP diversification, Functional genomics (DMS/CRISPR), and AI/ML-driven de novo design (e.g., MaSIF), including CRBN minimal degron and covalent MGs targeting DCAF16.
pubs.acs.org/doi/10.1021/...

04.02.2026 00:24 — 👍 0    🔁 0    💬 0    📌 0

On this Bluesky account,
I share bioinformatics explanations in English,
focused on protein structure prediction
and protein design. Here, I share technical insights for an international audience.

This differs from my Japanese posts on X
and is aimed at an international audience.

04.02.2026 00:12 — 👍 2    🔁 0    💬 0    📌 0