bioinformatics

The paper introduces LibStructBench, a document-grounded benchmark evaluating AI agents' ability to reconstruct sequencing library structures and molecular workflows, finding that frontier agents remain unreliable at exact reconstruction and that model choice matters more than agent harness choice.

该论文研究在769维特征下，完整人类基因评分数据集与激酶基因子集是否呈现相同的聚类结构，通过四种聚类方法比较发现最优聚类数均为两类。

The Consortium for Biomedical Research and AI in Neurodegeneration (C-BRAIN) is working with NVIDIA to use AI, federated learning, and GPU infrastructure to unlock untapped Alzheimer's research data and accelerate the pace of scientific discovery for the 57 million people living with dementia.

HuggingFaceBio 发布开源 14.8B DNA 理解与生成模型 Carbon-X，基于 Qwen3-14B 微调，融合 DNA 序列与自然语言训练，在五项 DNA 基准上平均得分 66.7%，接近前沿闭源模型但推理成本约为其 1/1800。

Hugging Face released Carbon-X, an open DNA language model that answers questions about gene sequences in plain English, scoring 66.7% on DNA benchmarks near GPT-6.1 Sol at roughly 1/1,800 the cost per problem. It runs on a single 80GB GPU, potentially making large-scale variant classification dramatically cheaper for hospitals and researchers, though it is not yet clinically validated.

Carbon-X is an open 14.8B decoder-only model that jointly processes raw DNA and text via a shared 6-mer token stream, achieving 66.7% average on five DNA benchmarks — nearly matching GPT-6.1 Sol — with Apache 2.0 weights that fit on a single 80GB GPU.

HuggingFace Bio 发布 Carbon-X，一个 14B 的 DNA+自然语言联合模型，在五项 DNA 基准上超越所有开源模型并匹配 GPT-6.1 Sol，且成本低约 1000 倍，同时提供 GGUF 版本支持 llama.cpp 本地运行。

Hugging Face released Carbon-X, an open model that lets users chat with DNA in natural language, trained on 300 billion tokens of genomic sequences and English text to help researchers interpret genetic mutations and design therapies.

提出了GOLD框架，通过GeoVAE将RNA离散序列与连续侧链角度投影到统一潜在流形上进行协同扩散，并引入训练无关的梯度引导采样机制，实现结构感知的RNA逆折叠，达到最先进性能。

Wired/Quanta 报道指出，AI 基因组基础模型（如 Evo 2、Genos、AlphaGenome）仍无法解读人类基因组复杂的三维物理结构及其基因调控机制，说明 AI 目前尚不足以真正破解我们的基因组。

Show HN: Grimoire for Biology — a new developer tool or platform aimed at biology, shared with the Hacker News community for feedback.

OmicsBench introduces a long-horizon, multi-turn benchmark of 97 DAG-structured bioinformatics tasks (662 subtasks) for evaluating AI agents over research-study-scale workflows, revealing significant performance gaps and introducing an unsupervised failure mode discovery pipeline.

Google DeepMind 展示如何利用其 agentic 开发平台 Google Antigravity 加速抗菌肽（AMP）的发现流程，结合 ProtNLM 蛋白质功能预测模型从 UniProt 未注释序列中筛选出新型抗菌肽候选物，目前正在湿实验验证阶段。

该论文对九种主流类增量学习（CIL）方法在五个冻结的蛋白质编码器上的表现进行了系统基准测试，发现没有方法能可靠跨编码器迁移，并分析了失败源于嵌入几何异质性与不稳定的梯度稳定性信号，主张持续学习设计应将表征本身作为核心组件。

Pythia, an interpretable AI model trained on a new high-resolution atlas of 67 RNA-binding proteins, is introduced as a collaboration between Genophoria, CZ Biohub, UCSF, and Calico to predict how RBPs recognize RNA by sequence and shape.

DeepMind has released AlphaGenome, a machine learning model that predicts the impact of genomic variants at base-pair resolution with a context window of one million base pairs. It matches or outperforms the strongest competing models on 25 of 26 variant effect benchmarks. Accompanied by the AlphaGenome Atlas, which has pre-computed the effects of roughly nine billion single-nucleotide variants — a scale 30 times that of the AlphaFold database — the model weights are permanently free for the scientific community.

Arc Institute argues that virtual cell models can only predict how cell types respond to perturbations if they are trained on high-quality causal data across diverse contexts, so technical noise is not mistaken for real biology.

ReLaG is a scalable, modality-agnostic framework that infers groups of related samples via proximity graphs and community detection to produce independent train–test splits, addressing the over-optimistic generalization estimates caused by random splits on data with latent relations. It scales better than existing relation-aware methods and offers a label-free procedure to adapt splitting resolution to production settings, available as pip-installable open-source software.

Forbes features ChronicleBio, founded by a former OpenAI executive, which is building a deep biological dataset and using AI to identify treatments for chronic conditions like POTS, Long COVID, ME/CFS, and hEDS that currently have zero FDA-approved therapies.

GATE-ST proposes a gene-aware text-image encoder that integrates text-encoded gene summaries with histology image embeddings via cross-attention to improve spatial gene expression prediction, outperforming image-only and random-embedding baselines in pathology imaging benchmarks.