CrisprDNT

Transformer-based Anti-noise Model for Predicting CRISPR-Cas9 Off-Target Activities.

PREREQUISITE

CrisprDNT was conducted by TensorFlow version 2.3.2 and python version 3.6.

Following Python packages should be installed:

• numpy

• pandas

• scikit-learn

• TensorFlow

• Keras

Usage

For a new dataset：

• You can process the dataset by calling code/data_process/create_coding_scheme.py to get the coding format needed by CrisprDNT.
• Once you have the coding format for the new dataset, you can train and test the dataset using code/train&test/experiment.py, which contains the training procedure for CrisprDNT. The model architecture for CrisprDNT is described in detail in code/model/model_network.py, which also contains other models for off-target prediction.

Data Description:

• dataset1->Doench et al.(Protein knockout detection)

• dataset2->Haeussler et al.(PCR, Digenome-Seq and HTGTS)

• dataset3->Cameron et al.(SITE-Seq)

• dataset4->Tasi et al.(GUIDE-seq)

• dataset5->Kleinstiver et al(GUIDE-seq)

• dataset6->Listgarten et al.(GUIDE-seq)

• dataset7->Chuai et al.(GUIDE-Seq, Digenome-Seq,HTGTS,BLESS and IDLVs)

• dataset8->Chuai et al.(GUIDE-Seq, Digenome-Seq,HTGTS,BLESS and IDLVs)

• CIRCLE_seq_10gRNA_wholeDataset->mismatches and indels

• elevation_6gRNA_wholeDataset->miamatches and indels

• Code Description

  • data_process(coding scheme)
    • create_coding_scheme.py->Create CrisprDNT, CRISPR_IP, CRISPR_Net and CNN_std encoding.
    • CnnCrispr_coding.py->Create CnnCrispr and CNN_std encoding.
    • CRISPR-OFFT_coding.py->Create CRISPR-OFFT encoding.
  • data_analyse
    • analyse_mismatchposition&type.py->Analysis of the effect of mismatch position&type based on GUIDE-seq dataset.
    • analyse_mismatchposition.py->Analysis of the effect of mismatch position based on GUIDE-seq dataset.
    • analyse_type.py->Analysis of the effect of mismatch type based on GUIDE-seq dataset.
  • model
    • model_network.py->CrisprDNT network and anti-noise loss function code.
  • train&test
    • experiment.py->code to reproduce the experiments with CrisprDNT, CRISPR_IP, CRISPR_Net, CnnCrispr, CRISPR-OFFT and CNN_std

saved_model Description:

• encodedmismatchtype14x23Kleinstiver_5gRNAwithoutTsai.pkl+new_crispr_ip.h5->CrisprDNT trained in dataset5.
• encodedmismatchtype14x23Kleinstiver_5gRNAwithoutTsai.pkl+NCEandRCE_new_crispr_ip.h5->CrisprDNT with NCE&RCE loss function trained in dataset5.
• encodedmismatchtype14x23Kleinstiver_5gRNAwithoutTsai.pkl+NCEandMAE_new_crispr_ip.h5->CrisprDNT with NCE&MAE loss function trained in dataset5.
• encodedmismatchtype14x23Kleinstiver_5gRNAwithoutTsai.pkl+sce_new_crispr_ip.h5->CrisprDNT with SCE loss function trained in dataset5.
• encodedmismatchtype14x23Kleinstiver_5gRNAwithoutTsai.pkl+gce_new_crispr_ip.h5->CrisprDNT with GCE loss function trained in dataset5.
• encodedmismatchtype14x23Listgarten_22gRNAwithoutTsai.pkl+new_crispr_ip.h5->CrisprDNT trained in dataset6.
• encodedmismatchtype14x23Listgarten_22gRNAwithoutTsai.pkl+NCEandRCE_new_crispr_ip.h5->CrisprDNT with NCE&RCE loss function trained in dataset6.
• encodedmismatchtype14x23Listgarten_22gRNAwithoutTsai.pkl+NCEandMAE_new_crispr_ip.h5->CrisprDNT with NCE&MAE loss function trained in dataset6.
• encodedmismatchtype14x23Listgarten_22gRNAwithoutTsai.pkl+sce_new_crispr_ip.h5->CrisprDNT with SCE loss function trained in dataset6.
• encodedmismatchtype14x23Listgarten_22gRNAwithoutTsai.pkl+gce_new_crispr_ip.h5->CrisprDNT with GCE loss function trained in dataset6.