Algorithm 2.

Proposed GSAF for PTM selection.

Step 1 Input: Training set $Y_1$ and validation set $Y_2$

Step 2 for k = 1: $N_{PTM}$ (k is the index of PTM)

for $L=1:L_{\text{max}}$ (L is the index of NLR)

Step 2.1 PTM Retrain

Import $k$-th PTM $M_0\left(k\right)$,

Use L2TFL via data $Y_1$ and removing $L$ layers,

Obtain $M_3\left(k,L\right)$,

Step 2.2 Feature Extraction

Generate features $f_M\left(k,L\right)$ from $M_3\left(k,L\right)$.

Step 2.3 Train OHNN

Initialize OHNN $B^i\left(k,L\right)$,

Train OHNN $B^i\left(k,L\right)$ using input as $f_M\left(k,L\right)$,

Obtain $B^t\left(k,L\right)$,

Step 2.4 Obtain Indicator

Obtain performance indicator $I\left(k,L\right)$ over validation set $Y_2$

end

end

Step 3 Generate and sort the indicator vector $\overrightarrow{I\left(k,L\right)},$

Step 4 Obtain the rank list $\overrightarrow{R_{GSAF}}$,

Step 5 Choose the top two best models (determine PTM and NLR):

$M\left[\overrightarrow{R_{GSAF}}\left(1\right)\right]$ and $M\left[\overrightarrow{R_{GSAF}}\left(2\right)\right]$