Tag
该论文发现 ECG 到 PPG 的跨模态蒸馏中，冻结 ECG 教师目标的 40–59% 是与个体身份相关的 per-recording 均值（trait），学生模型往往记住身份而非生理状态；作者提出 fixed-effects distillation，通过减去每条记录的均值使 trait 精确抵消，使 within-person 状态一致性提升一倍以上，并在两个骨干网络和两个额外数据库上验证了效果。
OpenAI Foundation funds initiatives to create high-quality datasets for AI in biology and medicine, aiming to overcome data limitations that hinder medical breakthroughs.
This study evaluates the use of Retrieval-Augmented Generation (RAG) and Named Entity Recognition (NER) to improve automated lay summaries of radiological reports, finding that NER enhances readability and quality, while RAG has variable effects depending on fine-tuning.
SSE-Bio proposes a structured self-evolving agent with a trainable retrieval policy to improve multi-hop biomedical reasoning, demonstrating significant performance gains over baselines.
This paper evaluates quantum-classical hybrid machine learning for lung cancer detection using cfDNA fragmentomics and methylation data, showing competitive performance of quantum kernel models against classical baselines.
The paper proposes P2E-VQ, a retrieval-augmented framework that enhances PPG representations by retrieving ECG-linked information via discrete patch retrieval, improving downstream tasks without requiring ECG during inference.
CADENCE uses sparse autoencoders to decompose ECG foundation model representations into interpretable physiological concepts, significantly improving alignment with clinical phenotypes and waveform morphology.
Omni-Sleep is a sleep foundation model that uses hierarchical contrastive learning to capture CNS-ANS dynamics from multimodal polysomnography signals, outperforming strong baselines on sleep staging and multi-disease classification.
MKGR is a multimodal framework that combines protein sequence encoding with four biomedical knowledge graphs to improve cold-start protein-protein interaction prediction, outperforming baselines on benchmark datasets.
This paper introduces an AI agent trained via reinforcement learning to reason over all FDA-approved drugs since 1939 for treatment recommendations, integrating disease context, comorbidities, and contraindications.
KG-TRACE is a neuro-symbolic framework that integrates a WHO mutation knowledge graph with a neural genomic model for antimicrobial resistance prediction, achieving high accuracy and introducing a Biological Grounding Ratio metric to ensure alignment with established biological knowledge.
BioManus is an MCP-native biomedical agent system that uses graph-scaffolded planning over structured biological capabilities instead of flat prompt-based tool retrieval, achieving better context efficiency and execution accuracy on biomedical benchmarks. The system introduces a BioinfoMCP Compiler to standardize heterogeneous bioinformatics tools and organizes them as a typed heterogeneous MCP graph for scalable reasoning.
GiG is a knowledge graph-modulated deep learning framework that integrates biological knowledge graphs as edges and patient-specific data as node features, outperforming SOTA by up to 49% in limited-sample clinical tasks.
This paper demonstrates that switching from Masked Language Modeling to Causal Language Modeling during encoder adaptation improves downstream performance on biomedical texts. The authors release ModernBERT-bio and ModernCamemBERT-bio as state-of-the-art biomedical encoders.
MIT released FINGERS-7B, a 7-billion-parameter multi-omics foundation model trained on data from 30,000 individuals to predict Alzheimer's risk years in advance. The model is accessible via the AD Workbench and is accompanied by a research paper on OpenReview.
This paper introduces NATD-GSSL, a framework evaluating the robustness of Graph Self-Supervised Learning on noisy, text-driven biomedical graphs. It demonstrates that certain GNN architectures and pretext tasks maintain performance despite real-world noise, offering practical guidance for unsupervised learning in imperfect datasets.
Researchers fine-tuned BioMistral-7B with QLoRA and GraphRAG to create a TB-care LLM for South Africa, showing improved contextual alignment over the base model.
LogosKG introduces a hardware-aligned framework for scalable, interpretable multi-hop retrieval on billion-edge knowledge graphs, integrating degree-aware partitioning and on-demand caching to boost efficiency without sacrificing fidelity.