KAN-GIN: Adaptive Nonlinear Molecular Representation Learning for Drug-Target Affinity Prediction

. 2026 Sep 14;19(9):1458. doi: 10.3390/ph19091458

Algorithm 1 KAN-GIN forward propagation for drug-target affinity prediction

Require:
Molecular graph $G = (V, E)$ with initial node representations ${h_{v}^{(0)}}_{v \in V}$ ; protein sequence $P = (a_{1}, a_{2}, \dots, a_{n})$ ; number of graph layers K
Ensure:
Predicted drug-target affinity $\hat{y}$
1:
for $k = 1$ to K do
2:
for each node $v \in V$ do
3:
$h_{v}^{(k)} \leftarrow {KAN}^{(k)} ((1 + ϵ^{(k)}) h_{v}^{(k - 1)} + \sum_{u \in N (v)} h_{u}^{(k - 1)})$
4:
end for
5:
end for
6:
$h_{G} \leftarrow max_{v \in V} h_{v}^{(K)}$
7:
$h_{D} \leftarrow h_{G}$
8:
$h_{P} \leftarrow ProteinEncoder (P)$
9:
$h \leftarrow [h_{D}; h_{P}]$
10:
$\hat{y} \leftarrow f (h)$
11:
return $\hat{y}$