An Augmented Sample Selection Framework for Prediction of Anticancer Peptides

. 2023 Sep 18;28(18):6680. doi: 10.3390/molecules28186680

Algorithm 1 Uncertainty-aware augmented sample selection (ACPs-ASSF)
Require: Original training dataset $D_{o}$ ; augmented sample dataset $D_{a}$ ; prediction model $f_{θ}$ with trainable parameters $θ$ ; uncertainty threshold $λ$ and confidence threshold $γ$ ; number of stochastic forward pass times $T$ ; number of iterations for selecting samples $I$ ; number of epochs $E$ for training model.
1:	$D = D_{o}$ ;	▷ obtain training set
2:	for $i t e r a t i o n = 1$ to $I$ do
3:	Initialize $θ$ ;
4:	if $i t e r a t i o n > 1$
5:	$D = D_{o} \cup D_{s}$ ;	▷ merge the selected samples to the training set
6:	for $e p o c h = 1$ to $E$ do
7:	Train $f_{θ}$ using $D$ ;	▷ using CE loss and SGD
8:	end for
9:	for $t = 1$ to $T$ do
10:	$D r o p o u t (θ)$ ;
11:	Input samples from $D_{a}$ into $f_{θ}$ ;	▷ accumulate the output of each pass
12:	end for
13:	Compute the uncertainty and pseudo-labels by Equations (9) and (10);
14:	Use Equation (11) to obtain $D_{s}$ ;	▷ select augmented samples
15:	end for
16:	return $θ$