Skip to main content

Table 3 Performance comparison with states-of-the-art models under test pattern 1

From: Deep learning improves the ability of sgRNA off-target propensity prediction

Test Set

Model

auROC

auPRC

Pearson value

Spearman value

Total test set

CnnCrispr

0.975

0.679

0.682

0.154

CFD

0.942

0.316

0.343

0.140

MIT

0.77

0.044

0.150

0.085

CNN_std

0.947

0.208

0.321

0.141

DeepCrispr

0.981

0.497

0.133

Hek293t test set

CnnCrispr

0.971

0.686

0.712

0.160

CFD

0.936

0.318

0.371

0.143

MIT

0.756

0.048

0.153

0.084

CNN_std

0.939

0.204

0.330

0.144

DeepCrispr

0.984

0.521

0.136

K562 test set

CnnCrispr

0.995

0.688

0.426

0.134

CFD

0.965

0.322

0.336

0.128

MIT

0.814

0.033

0.057

0.086

CNN_std

0.983

0.287

0.319

0.132

DeepCrispr

0.953

0.41

0.126

  1. We downloaded the prediction models of CFD, MIT and CNN_std from relevant websites and obtained the prediction results on the same test set as CnnCrispr. Since the training process of CnnCrispr was consistent with DeepCrispr’s, we directly used the test results in Additional file 2 given by DeepCrispr for performance comparison. The numbers in boldface indicate the highest scores for each metric