Tag
New work shows that genome language models like Evo 2 can perform in-context learning from a few examples, generating proteins, de novo antibodies, CRISPR-Cas9 loci, mitochondrial genomes, and even complete bacteriophages, with iterative chain-of-thought-style prompting for harder design goals.
A large China-based collaboration released Pinal, an AI tool that turns natural language prompts into functional protein designs, while Hugging Face launched a similar tool for DNA-level genome design—bringing 'vibe coding' style accessibility to protein and genome engineering.
Anthropic and Citrini Research used an internal Claude model to autonomously generate protein minibinders against five scorpion neurotoxins, yielding 17 functional binders from 211 tested, marking a step toward AI-driven antivenom discovery.
Artios is released as a fully open source system for designing functional de novo enzymes, aiming to enable precise coordination of chemistry in protein design and accelerate biology engineering in the open.
一支几乎没有生物学背景的五人投资团队利用 Claude 协调 300 多个 AI 智能体,在 64 小时内设计了 318,484 种蛋白质,其中 17 种经独立实验室验证可结合危险的蝎子神经毒素。
Riccardo De Santi et al. introduce LLM-to-Flow Distillation (LFD), a method that lets an LLM grade a scientific generator's outputs against plain-English goals, removing the need for task-specific reward models and outperforming direct reward-function training in drug/molecule design.
本文提出 PIDiff,一种联合离散序列空间与 SE(3) 结构流形扩散的多模态框架,用于蛋白质-蛋白质复合物设计,并提出 PIFormer 架构与 PIBench 数据集,在 minibinder 和抗体设计任务上超越基线方法。
论文提出 DRYLAB-Bench,首个评估 LLM 智能体高风险生物突变设计能力的综合基准,通过 30 个危险突变设计场景和 187,695 条湿实验数据量化生物安全风险;对 15 个前沿模型的评估显示生物设计工具会显著放大风险。
研究人员利用生成式AI设计蛋白质,并从31,565个基因组中筛选出数千个PiggyBac转座酶家族成员,其中一个AI设计变体使培养细胞中的靶向DNA插入效率翻倍,扩展了基因编辑工具集。
Google DeepMind and collaborators present AlphaProtein Novo, a machine-learning pipeline for generative de novo enzyme design that produces new-to-nature enzymes with state-of-the-art activity on benchmark reactions, including enzymes for synthesizing piperidine and degrading the environmental toxin DEHP.
Google DeepMind published a research paper on a protein watermarking system, adapting SynthID-style probability-biasing watermarks to AI-designed protein sequences so that trusted research proteins can be identified, helping mitigate biosecurity risks from AI protein design tools.
Google DeepMind announced SynthID Bio, a watermarking technology that embeds verifiable signatures into AI-designed proteins while preserving their biological function, verified in wet-lab tests and built into AlphaFold 3's diffusion network.
论文提出 Protein-TetSphere,一种将蛋白链四面体化并注册到固定拓扑参考体积、以共享拉普拉斯基表示的逐残基体积表征,用于配体结合口袋分类、蛋白-蛋白界面预测和 de novo binder 设计,在三项任务上均取得显著提升。
Mirror-Score introduces a calibrated, inference-only scoring framework for D-peptide/L-protein binder design, releasing the first public benchmark of 31 crystal complexes and showing that raw ProteinMPNN NLL is a poor affinity ranker (Spearman 0.19), while Boltz-2 interface pLDDT achieves structure-level leave-one-out Spearman 0.90 on the viral-entry family. The paper also reports that cross-family calibration fails to transfer and proposes a prospective design protocol for antimicrobial-resistance targets LasR and LecB.
The article reports on AI BioDesign, a project using artificial intelligence and experiments to create molecules not found in nature, led by Nobel laureate David Baker, aiming to advance medicine and technology.
Adaptyv Bio and Anthropic have partnered to launch a global Protein Design Competition, using AI to design new drug candidates with over $2 million in support and open data sharing.
Anthropic optimized inference for over 30 open-source biomolecular models using Claude, achieving an average 4x speedup, and is open-sourcing the optimization code. They are also partnering with Adaptyv Bio on a protein design competition.
OpenAI has shut down a protein design project, prompting the user to turn to open weight models for biological research.
ABLE is a benchmark for evaluating LLM agents' ability to use biological AI models like ProteinMPNN and AlphaFold3 in protein design workflows. It assesses 15 frontier models, finding that Claude Sonnet 4 and Gemini 3 Pro achieve the highest scores, while some models refuse all tasks.
GPT-6 Astra, an AI model, is viral for its diverse applications in gaming, protein design, sports tracking, and 3D world building, highlighted by 10 stunning cases.