Not that all the computation are scaled by the genome size and not at the read number as in [Hu et al.](https://doi.org/10.1093/nar/gkv670), this is also why we add a scaling factor (default to $10^3$).
Not that all the computation are scaled by the genome size and not at the read number as in [Hu et al.](https://doi.org/10.1093/nar/gkv670), this is also why we add a scaling factor $\beta$ (default to $10^3$).
The $\text{ratio}IP\left(t\right)$ is multiplied by this scaling factor.
The $\text{ratio}IP\left(t\right)$ is multiplied by this scaling factor.
With this method, we retain the interesting properties of [Hu et al.](https://doi.org/10.1093/nar/gkv670) normalization on the average read density between samples (i.e., we can compare two different samples in a quantitative way) and we account for the local bias of read density observed in the WCE samples (differential chromatin accessibility, repetition, low mappability region, etc.).
With this method, we retain the interesting properties of [Hu et al.](https://doi.org/10.1093/nar/gkv670) normalization on the average read density between samples (i.e., we can compare two different samples in a quantitative way) and we account for the local bias of read density observed in the WCE samples (differential chromatin accessibility, repetition, low mappability region, etc.).