Table 2.

Mean squared error (MSE) of the top five one-feature and two-feature combinations of the linear regression, support vector regression (SVR) and k-nearest neighbors regression (KNN) models for predicting aggregation rates. There are 20 mAbs in this study. The MSE are averaged from 100 randomly generated fourfold cross-validation sets

	One-feature	MSE	Two-features		MSE
	SCM_neg_H2	5.04	SCM_neg_H2	SASA_phobic_H3	4.81
	SAP_pos_H1	5.31	SCM_neg_H2	SASA_philic_L3	4.97
Linear	SASA_phobic_H3	5.49	SAP_pos_L1	SCM_neg_H2	5.08
	SCM_neg_H1	5.66	SCM_neg_H1	SASA_phobic_H3	5.19
	SASA_philic_L3	5.70	SCM_neg_H2	SCM_pos_L1	5.23
	SCM_pos_H2	4.96	SCM_pos_H2	SASA_phobic_Fv	4.12
	SCM_neg_H2	5.14	SAP_pos_L1	SCM_pos_H2	4.68
SVR	SCM_pos_L3	5.43	SAP_pos_L1	SCM_neg_H2	4.89
	SASA_phobic_Fv	5.44	SAP_pos_Fv	SASA_phobic_Fv	4.90
	SAP_pos_L1	5.46	SCM_pos_H2	SCM_pos_L3	4.90
	SCM_pos_H2	4.35	SCM_pos_H2	SASA_phobic_Fv	3.37
	SCM_pos_L3	4.97	SAP_pos_L1	SCM_pos_H2	3.80
KNN	SCM_neg_H1	5.35	SCM_neg_H1	SCM_pos_H2	3.97
	SCM_pos_H1	5.59	SCM_pos_H2	SASA_philic_L3	4.21
	SAP_pos_Fv	5.65	SCM_pos_L3	SASA_philic_L1	4.73