发表机构
Georgia Institute of Technology; Parker H. Petit Center for AI-Driven Health Innovation（佐治亚理工学院; 帕克·H·佩蒂特人工智能驱动健康创新中心）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本研究通过因果干预和探测揭示AlphaFold 3利用MSA作为结构捷径，在早期块形成全局几何，且比对可跨蛋白质转移折叠信息，提升预测精度。
AI 中文摘要
AlphaFold 3 以惊人的准确性预测蛋白质结构，然而结构信息如何在模型内部涌现仍鲜为人知。在此，通过对内部表征的因果干预以及对每个 Pairformer 块的直接探测，我们追溯了全局蛋白质几何结构的形成过程，并识别出多序列比对（MSA）作为折叠的结构捷径。移除 MSA 在很大程度上保留了局部二级结构，同时破坏了定义全局拓扑的长程关系。仅在四十个残基处恢复 MSA 富集的配对表征，即可恢复大部分丢失的组织，包括在从未直接修改的配对处。这种贡献依赖于 MSA 模块输出的详细方向而非其幅度。Pairformer 迅速将该信号转化为全局几何结构：对于大多数蛋白质，最终折叠在约 48 个块中的第 9 个块处即可恢复，比模型解码器能够渲染它早约 27 个块，而若无 MSA，则在遍历过程中对大多数蛋白质仍不可及。提供哪些同源物比提供哪些查询更能塑造这一轨迹；对于从未进化过的设计查询，它持续存在，但对于打乱的序列则崩溃。最重要的是，为共享折叠的不同蛋白质构建的比对，仅在结构对应的列处提供，将实验的中位 TM-score 从 0.44 提高到 0.72，而相同的比对沿链移动几个残基则比不提供任何比对表现更差。因此，AlphaFold 3 从比对中读取的是折叠本身的描述，可在共享折叠的蛋白质之间转移，而非查询自身的进化历史。这解释了其准确性及其已解决问题之局限。
英文摘要
AlphaFold 3 predicts protein structures with remarkable accuracy, yet how structural information emerges within the model remains poorly understood. Here, through causal interventions on internal representations and direct probing of every Pairformer block, we trace the formation of global protein geometry and identify the multiple sequence alignment (MSA) as a structural shortcut to the fold. Removing the MSA largely preserves local secondary structure while disrupting the long-range relationships that define global topology. Restoring the MSA-enriched pair representation at only forty residues recovers most of this lost organization, including at pairs never directly modified. This contribution depends on the detailed direction of the MSA module's output rather than its magnitude. The Pairformer rapidly converts this signal into global geometry: the final fold becomes recoverable by approximately block 9 of 48 for a majority of proteins, roughly twenty-seven blocks before the model's decoder can render it, whereas without the MSA it remains inaccessible for most proteins throughout the pass. Which homologs are supplied shapes this trajectory more strongly than which query is supplied; it persists for a designed query that never evolved but collapses for a shuffled sequence. Most importantly, an alignment built for a different protein that shares the fold, supplied only at the structurally corresponding columns, raises the median TM-score against experiment from 0.44 to 0.72, while the same alignment shifted a few residues along the chain performs worse than supplying no alignment at all. What AlphaFold 3 reads from an alignment is therefore a description of the fold itself, transferable between proteins that share one, rather than the query's own evolutionary history. This explains both its accuracy and the limits of what it has solved.
