Table 1 Performance of SPOT-RNA on validation and test set after initial training, transfer learning, and direct training.

From: RNA secondary structure prediction using an ensemble of two-dimensional deep neural networks and transfer learning

Method

Training set

Analysis set

MCC\({}^{a}\)

F1\({}^{b}\)

Precision

Sensitivity

Initial training

TR0

VL0

0.632

0.629

0.712

0.563

 

TR0

TS0

0.629

0.626

0.709

0.560

 

TR0

TS1

0.650

0.630

0.897

0.485

Transfer learning

TR1+VL1

TR1+VL1

0.701 (0.02\({}^{c}\))

0.690 (0.02\({}^{c}\))

0.853 (0.02\({}^{c}\))

0.580 (0.03\({}^{c}\))

 

TR1+VL1

TS1

0.690 (0.02\({}^{c}\))

0.687 (0.01\({}^{c}\))

0.888 (0.02\({}^{c}\))

0.562 (0.02\({}^{c}\))

Direct training

TR1

VL1

0.583

0.546

0.854

0.401

 

TR1

TS1

0.571

0.527

0.870

0.378

  1. aMatthews correlation coefficient
  2. bHarmonic mean of precision and sensitivity
  3. cStandard deviation based on five-fold cross-validation
Back to article page