SMPLIP-Score: predicting ligand binding affinity from simple and interpretable on-the-fly interaction fingerprint pattern descriptors

Table 4 Complexity comparisons of SMPLIP-Score

Models	Features	No of features	Learning parameters	Run time (s)^a	Complexity^a	Memory usage (max)^b
SMPLIP-RF	IFP	140	max_features (dʹ) = ‘auto’ or ‘sqrt’ n-estimators (t) = 100 to 500	2.82	\(O(tO(d{^{\prime}}n{^{\prime}}*{\log}(n{^{\prime}}))\)	55.24
	IFP + Int-Dist	280		6.55		56.621
	IFP + Frag	2422		36.86		57.18
	IFP + Int-Dist + Frag	2562		40.37		57.17
SMPLIP-DNN	IFP	140	159,601	137.96	\(O(d\mathrm{n}layers*nodes)\)	130.49
	IFP + Int-Dist	280	215,601	161.83		131.73
	IFP + Frag	2422	1,072,401	295.79		130.93
	IFP + Int-Dist + Frag	2562	1,128,401	495.57		131.18
SMPLIP-Linear	IFP + Frag	2422	Loss = huber, penalty = elasticnet, max_iter (\({t}_{i})\)=100	1.54	\(O({t}_{i}dn)\)	57.66
PLEC-Linear^c	PLEC FP	65,536	Loss = huber, penalty = elasticnet, max_iter (\({t}_{i})\)=100	–	\(O({t}_{i}dn)\)	–

^a,b The run time and memory usages were computed on system (Intel Xeon CPU E5-2650) using PDBbindv.2015-refined set
^aThe comparison of time complexity according to chosen features or learning condition
^bThe comparison of space complexity according to chosen features or learning condition
^cThe data was gained from the original article of PLEC [32]

ISSN: 1758-2946