Hi-C, split-pool recognition of interactions by tag extension (SPRITE) and genome architecture mapping (GAM) are powerful technologies utilized to probe chromatin interactions genome wide, but how faithfully they capture three-dimensional (3D) contacts and how they perform relative to each other is unclear, as no benchmark exists. Here, we compare these methods in silico in a simplified, yet controlled, framework against known 3D structures of polymer models of murine and human loci, which can recapitulate Hi-C, GAM and SPRITE experiments and multiplexed fluorescence in situ hybridization (FISH) single-molecule conformations. We find that in silico Hi-C, GAM and SPRITE bulk data are faithful to the reference 3D structures whereas single-cell data reflect strong variability among single molecules. The minimal number of cells required in replicate experiments to return statistically similar contacts is different across the technologies, being lowest in SPRITE and highest in GAM under the same conditions. Noise-to-signal levels follow an inverse power law with detection efficiency and grow with genomic distance differently among the three methods, being lowest in GAM for genomic separations >1 Mb.

Hi-C、通过标签扩展 (SPRITE) 和基因组结构映射 (GAM) 进行相互作用的分离池识别是用于探测全基因组染色质相互作用的强大技术，但它们如何忠实地捕获三维 (3D) 接触以及它们如何执行相对于彼此尚不清楚，因为不存在基准。在这里，我们将这些方法在一个简化但受控的框架中与小鼠和人类基因座聚合物模型的已知 3D 结构进行比较，该框架可以概括 Hi-C、GAM 和 SPRITE 实验以及多重荧光原位杂交 (FISH) 单-分子构象。我们发现在 silico Hi-C、GAM 和 SPRITE 批量数据中忠实于参考 3D 结构，而单细胞数据反映了单分子之间的强烈可变性。在重复实验中返回统计上相似的联系人所需的最小单元数在不同的技术中是不同的，在相同条件下，SPRITE 中最低，GAM 中最高。噪声信号水平遵循具有检测效率的反幂律，并且在三种方法中随着基因组距离的不同增长，在基因组分离 >1 Mb 的 GAM 中最低。