Computational protocol: Impact of data resolution on three-dimensional structure inference methods

Similar protocols

Protocol publication

[…] The first model we consider is the No Random Effect (NRE) model: logλij=β0+β1logdij+γ1log(zl,izl,j)+γ2log(zg,izg,j)+log(zm,izm,j), where the coefficient β1 was set to be −0.434, which was the estimate obtained along with the gold standard structure ωgs []. Note that this parameter is far from −1 used in many optimization-based methods. For (γ1,γ2), three sets of values are considered to entertain a variety of potential covariates: (0.3,0.3),(0.05,0.25), and (0.05,−0.25). Following the argument in [], we set β0=3 arbitrarily as its value only affects the scale of the predicted structure, thus not altering its correlation with genomic functions. Finally, to mimic real Hi-C data and guided by the ranges of data for three factors that can affect the counts [], we set zl,i∼Unif(0.2,0.3), zg,i∼Unif(0.4,0.5) and zm,i∼Unif(0.9,1), where Unif(.) denotes a uniform distribution. Note that this model coincides with the analysis model employed by tPAM, BACH, and PASTIS, and therefore these methods are expected to perform well, especially when there is no zero-inflation. [...] We compared tREX with its fellow modeling-based methods (BACH [], tPAM [] and PASTIS []) as well as optimization-based methods (ShRec3D [] and ChromSDE []). In fitting tREX, tPAM, and BACH to the simulation data, we normalized the data by incorporating the systematic bias information as covariates to the model. In contrast, we first normalized the data by HiCNorm [] before using ShRec3D, ChromSDE and PASTIS to the normalized data. In order to assess and compare estimation accuracy of the methods, we need to take into account the fact that the estimated architecture is accurate only up to a scaling factor. Therefore, we first estimate the scaling factor α by the least squares model as α^=argminα∑ ∑1≤i

Pipeline specifications

Software tools PASTIS, ShRec3D, ChromSDE, HiCNorm
Application Hi-C analysis
Organisms Homo sapiens