[论文] Screening of Biosecurity Features in Metagenomic Data with Evo 2 Probe...

论文概要

研究领域: ML 作者: Jeremy Guntoro, Alexander Dack, Dylan Danno, Michaela Jančovičová, Križan Jurinović, Vanessa Smilansky 发布时间: 2026-07-15 arXiv: 2607.14070

中文摘要

基因组基础模型如Evo 2学习丰富的序列表示，但其在生物安全筛选中的价值很大程度上未被探索。我们询问这些表示中有多少生物安全相关信号是线性可访问的，通过在冻结Evo 2第26层激活上训练最小线性和注意力探针，而不微调底层模型。在保留的宏基因组测试集上，探针以强判别力检测抗菌素耐药性（AMR）：线性探针达到区域级ROC-AUC 0.888（均值池），单头注意力探针提升至0.977。探针解析更细粒度的AMR药物类别子类别，并将它们与无关功能基因分离，提供额外证据表明学习信号不能仅由通用功能基因状态解释。细菌毒力也可解码，尽管较弱（区域级ROC-AUC 0.833）。AMR探针在模拟短读段上无需重新训练即可保持相当的排序性能，可在组装计算成本高或不可靠的环境中组装前评估。它达到读层级ROC-AUC 0.898（均值池），与均值池全区域结果相当。在SynGenome内，AMR相关提示标签仅从Evo 1.5生成的序列中弱恢复；这些提示派生标签不确立生成响应序列的功能。互补的稀疏自编码器分析恢复可解释的耐药性相关特征，但证明比监督探针一致性差。总之，这些结果将轻量级嵌入探针定位为宏基因组生物监测的快速、廉价首遍检测层，并绘制该方法的强项和当前局限。

原文摘要

Genomic foundation models such as Evo 2 learn rich sequence representations, but their value for biosecurity screening is largely unexplored. We ask how much biosecurity-relevant signal is linearly accessible in these representations by training minimal linear and attention probes on frozen Evo 2 layer-26 activations, without fine-tuning the underlying model. Across held-out metagenomic test sets, the probes detect antimicrobial resistance (AMR) with strong discrimination: a linear probe reaches a region-level ROC-AUC of 0.888 (mean-pool), rising to 0.977 with a single-head attention probe. The probes resolve finer-grained AMR drug-class subcategories and separate them from unrelated functional genes, providing additional evidence that the learned signal is not explained solely by generic ...

--- *自动采集于 2026-07-17*

#论文 #arXiv #ML #小凯