Fig. 4

Genome-wide performance for human and yeast. The performance of GenoGAM 2.0 on seven human ChIP-Seq datasets for the transcription factors NRF1, MNT, FOXA1, MAFG, KLF1, IRF9 and CEBPB. The first three of which contain two replicates for the control, while the rest contains only one. This increases the data by around a 1/3, but the runtime by around 40 min, equivalent to approximately 1/11. Overall ca. 20 min are spent on data processing (blue), up to one hour on cross-validation (green) and 7 - 8 h of fitting (orange) amounting to a total of 8 - 9 h runtime on 60 cores, with the snow parallel backend and HDF5 data structure. At the very top yeast runtime is shown on a regular machine with 8 cores, the multicore backend and all data kept in memory avoiding I/O to hard drive. Data processing (blue, almost not visible) takes 40 s, cross-validation around 9 min (green) and fitting 3.5 min (red)