This is an explanation of how to use DeepMod without examples. If you want to run some examples, please refer to demo.
The inputs of DeepMod is a group of FAST5 files and a reference genome. FAST5 files need to be basecalled already, and Events data must be availabe in FAST5 files.
DeepMod has a functional module called "detect" which will detect a specific modification in a single run. However, if the dataset and genome size is very larger or one wants to have the results soon, it would be better to run "detect" in a separate process simultaneously and then merge them together. For some special cases, if there is cluster effect between modifications (such as 5mC in CpG cluster), the third process would be used for additional prediction after "detect". How to use the three functional modules is described below.
The command for modification detection is to run python DeepMod.py detect. Without any other parameters, the help document will be shown. An example of how to use it is given below.
python DeepMod/bin/DeepMod.py detect --wrkBase FAST5-Folder --Ref Ref_genome_path --outFolder out_folder --Base C --modfile train_mod/rnn_f7_wd21_chr1to10_4/mod_train_f7_wd21_chr1to10 --FileID User_Uniq_name --threads 4
where users need to provide where is the FAST5 files (--wrkBase), where is the reference genome (--Ref), where is the output folder (--outFolder), and also the base of interest and the mod file. Users can optionally specify unique string for the results file names (--FileID) and how many threads are used (--threads).
If you want to make the prediction for base A, the following command could be used.
python DeepMod/bin/DeepMod.py detect --wrkBase FAST5-Folder --Ref Ref_genome_path --outFolder out_folder --Base A --modfile train_mod/rnn_conmodA_P100wd21_f7ne1u0_4/mod_train_conmodA_P100wd21_f3ne1u0 --FileID User_Uniq_name --threads 4
Some projects might generate very large Nanopore sequencing data. For example, NA12878 Nanopore sequencing data was ~30TB. To speed up the detection of modification, users can run DeepMod with different --FileID and folders where fast5 files are (--wrkBase) but the same output folder (--outFolder). Then, the following script can be used to merge modification detection grouped by chromosomes for human genome.
python DeepMod/tools/sum_chr_mod.py outFolder base-of-interest res-unique-filename chromosomes
The last parameter is optional if running on human genome; otherwise, the chromosomes should be provided by a string where chromosome names are seperated by ','. res-unique-filename is a unique prefix of the summary file under outFolder. outFolder is the output folder of DeepMod and the prediction of DeepMod must be in the sub-folders under outFolder.
5mC in CpG motifs has cluster effect in human genome. To consider cluster effect, a second deep learning process was designed to improve the 5mC detection performance. To do that, additional commands below are used
python DeepMod/tools/generate_motif_pos.py ref-genome result-folder C CG 0
The result files were generated under the directory of result-folder.
python DeepMod/tools/hm_cluster_predict.py prefix-merged-bed-files genome_motif_folder-in-last-step
The output files will be under the same directory of prefix-merged-bed-files but with the prefix of prefix-merged-bed-files by appending "_clusterCpG".