POSEIDON: Peptidic Objects SEquence-based Interaction with cellular DOmaiNs: a new database and predictor

Table 4 Results for the best performance of each optimized ML model

Model	Subset	RMSE	MSE	MAE	Pearson	Spearman	r²
Support vector machine	Train	0.550	0.303	0.318	0.918	0.943	0.817
Support vector machine	Test	0.717	0.514	0.485	0.856	0.887	0.706
Stochastic gradient descent	Train	–	–	–	0.002	0.002	–
Stochastic gradient descent	Test	–	–	–	0.002	0.002	–
k-nearest neighbors	Train	0.419	0.175	0.263	0.946	0.942	0.894
k-nearest neighbors	Test	0.746	0.557	0.572	0.828	0.823	0.681
Decision tree	Train	0.690	0.476	0.488	0.844	0.824	0.712
Decision tree	Test	0.951	0.904	0.669	0.704	0.707	0.483
Random forest	Train	0.397	0.158	0.259	0.958	0.961	0.905
Random forest	Test	0.700	0.490	0.452	0.856	0.869	0.720
Extreme randomized trees	Train	0.527	0.277	0.349	0.915	0.924	0.832
Extreme randomized trees	Test	0.765	0.585	0.522	0.819	0.830	0.665
Extreme gradient boosting	Train	0.177	0.031	0.098	0.991	0.989	0.981
Extreme gradient boosting	Test	0.643	0.413	0.394	0.874	0.881	0.764
Deep neural network	Train	0.259	0.067	0.153	0.980	0.979	0.959
Deep neural network	Test	0.640	0.410	0.402	0.876	0.880	0.765
Forked neural network	Train	0.358	0.128	0.199	0.961	0.960	0.923
Forked neural network	Test	0.740	0.547	0.447	0.839	0.857	0.687

ISSN: 1758-2946