How Private Are DNA Embeddings? Inverting Foundation Model Representations of Genomic Sequences
Methods in Medical Informatics, Department of Computer Science, University of Tuebingen, Germany（图宾根大学医学信息学方法系，计算机科学系）；Institute for Bioinformatics and Medical Informatics (IBMI), University of Tuebingen, Germany（图宾根大学生物信息学与医学信息学研究所）
DNA foundation models have become transformative tools in bioinformatics and healthcare applications. Trained on vast genomic datasets, these models can be used to generate sequence embeddings, dense vector representations that capture complex genomic information. These embeddings are increasingly being shared via Embeddings-as-a-Service (EaaS) frameworks to facilitate downstream tasks, while supposedly protecting the privacy of the underlying raw sequences. However, as this practice becomes more prevalent, the security of these representations is being called into question. This study evaluates the resilience of DNA foundation models to model inversion attacks, whereby adversaries attempt to reconstruct sensitive training data from model outputs. In our study, the model's output for reconstructing the DNA sequence is a zero-shot embedding, which is then fed to a decoder. We evaluated the privacy of three DNA foundation models: DNABERT-2, Evo 2, and Nucleotide Transformer v2 (NTv2). Our results show that per-token embeddings allow near-perfect sequence reconstruction across all models. For mean-pooled embeddings, reconstruction quality degrades as sequence length increases, though it remains substantially above random baselines. Evo 2 and NTv2 prove to be most vulnerable, especially for shorter sequences with reconstruction similarities > 90%, while DNABERT-2's BPE tokenization provides the greatest resilience. We found that the correlation between embedding similarity and sequence similarity was a key predictor of reconstruction success. Our findings emphasize the urgent need for privacy-aware design in genomic foundation models prior to their widespread deployment in EaaS settings. Training code, model weights and evaluation pipeline are released on: https://github.com/not-a-feature/DNA-Embedding-Inversion.
dnaGrinder: a lightweight and high-capacity genomic foundation model
Hong Kong Jockey Club STEM Laboratory of Genomics and AI in Healthcare（香港赛马会基因组学与医疗人工智能 STEM 实验室）；Department of Health Technology and Informatics（健康技术与信息学系）；Department of Data Science and Artificial Intelligence（数据科学与人工智能系）；Department of Computing（计算机系）；The Hong Kong Polytechnic University（香港理工大学）
JEPA-DNA: Grounding Genomic Foundation Models through Joint-Embedding Predictive Architectures
Applied AI Architecture, NVIDIA, Israel（NVIDIA应用人工智能架构，以色列）；Worldwide Field Ops, NVIDIA, Israel（NVIDIA全球现场运营，以色列）；Developer Programs, NVIDIA, Israel（NVIDIA开发者计划，以色列）；Cancer Research Center and Wohl Institute of Translational Medicine, Sheba Medical Center, Tel Hashomer, Israel（癌症研究中心和Wohl转化医学研究所，Sheba医疗中心，Tel Hashomer，以色列）；Windreich Department of AI and Human Health, Icahn School of Medicine at Mount Sinai, New York, USA（AI与人类健康风reich部门，Mount Sinai医学中心，纽约，美国）
Phenotypic divergence of Homo sapiens is driven by the evolution of human-specific genomic regulatory networks via two mechanistically distinct pathways of creation of divergent regulatory DNA sequences
Evaluating DNA function understanding in genomic language models using evolutionarily implausible sequences
Center for Interdisciplinary Studies, School of Science, Westlake University, Hangzhou, China（西湖大学交叉科学中心，理学院，杭州，中国）；Department of Quantitative and Computational Biology, University of Southern California, Los Angeles, CA, United States（定量与计算生物学系，南加州大学，洛杉矶，CA，美国）
Cancer-inspired Genomics Mapper Model for the Generation of Synthetic DNA Sequences with Desired Genomics Signatures
School of Mechanical Engineering, Tel Aviv University（特拉维夫大学机械工程学院）；Department of Cancer Biology, Cancer Institute, University College London（伦敦大学学院癌症研究所癌症生物学系）
Model Decides How to Tokenize: Adaptive DNA Sequence Tokenization with MxDNA
Shanghai Artificial Intelligence Laboratory（上海人工智能实验室）；Shanghai Jiao Tong University（上海交通大学）；The Chinese University of Hong Kong（香港中文大学）；The University of Sydney（悉尼大学）
本文提出 MxDNA 框架，通过稀疏卷积专家混合与可变形卷积使模型自主梯度学习 DNA 分词策略，考虑基因组片段特性，以更少数据时间超越现有方法并捕获基因组功能。
Highly conserved sequence-specific double-stranded DNA binding networks contributing to divergent genomic evolution of human and chimpanzee brain development
Institute of Engineering in Medicine University of California, San Diego（加州大学圣地亚哥分校工程医学院）
DART-Eval: A Comprehensive DNA Language Model Evaluation Benchmark on Regulatory DNA
Department of Computer Science, School of Engineering, Stanford University（计算机科学系，工程学院，斯坦福大学）；Department of Genetics, School of Medicine, Stanford University（遗传学系，医学院，斯坦福大学）；Department of Pathology, School of Medicine, Stanford University（病理学系，医学院，斯坦福大学）
Entropy, Disagreement, and the Limits of Foundation Models in Genomics
Genome Institute of Singapore, A*STAR（新加坡基因组研究院，A*STAR）；KU Leuven（卢森堡大学）；Faculty of Electrical Engineering and Computing, University of Zagreb（扎格雷布大学电子工程与计算学院）
Genome-Factory: A Library for Tuning, Deploying, and Interpreting Genomic Foundation Models
Center for Foundation Models；Generative AI \& Department of Computer Science, Northwestern University, USA；School of Natural Sciences, University of California at Merced, USA；Department of Energy Joint Genome Institute, Lawrence Berkeley National Laboratory, USA；Systems Biology Division, Lawrence Berkeley National Laboratory, USA；Department of Statistics；Data Science, Northwestern University, USA
Embedding Is (Almost) All You Need: Retrieval-Augmented Inference for Generalizable Genomic Prediction Tasks
Department of Computer Science and Engineering, Bangladesh University of Engineering and Technology（孟加拉工程与技术大学计算机科学与工程系）；Department of Computer Science and Engineering, BRAC University（布拉克大学计算机科学与工程系）
