A VGAE-based model to infer transcription factor regulatory network (TRN). TRN is represented as an undirected graph, where V nodes represent TFs and E edges represent their interactions. The input of DeepTFni is solely a scATAC-seq count matrix. The output of DeepTFni is the imputed TRN.
TRN is represented as an undirected graph, where V nodes represent TFs and
E edges represent their interactions. The input of DeepTFni is solely a scATAC-seq
count matrix. The output of DeepTFni is the imputed TRN. Taking TRN inference as a link prediction task, DeepTFni workflow is organized as follows:
DeepTFni is mostly written in Python 3.6 and some preprocessing steps are done in Perl 5. It can be run on a single desktop using Linux platform. To run DeepTFni, some prerequisites need to be installed. A detailed dependency list is:
- pytorch 1.7.1
- numpy 1.8.11
- pandas 1.1.3
- matplotlib 3.3.2
- MAESTRO 1.5.0
- BioPerl 1.5.7
- R 4.0.3
- fimo
Use DeepTFni through command:
perl run_DeepTFni_csv.pl Path_to_Input reference #Bash
For example:
perl run_DeepTFni_csv.pl ./Input_data/Input.csv hg19 #Bash
It will generate a new script based on run_template_csv.sh named like ‘run_your_input.sh’. The input should be organized like ./demo/scATAC_demo.csv, that is, each row represents a peak and each column represents a cell. If DeepTFni goes smoothly, it will finally return the imputed TRN represented by a symmetric adjacency matrix stored at ./train_info/result_k_10_summary.