Table 1.

Complexity and performance of 3F and Reference models on genotype-phenotype data sequenced at Virco up to September 2006

		Reference Sep 2006			3F Sep 2006 a			Unseen data
								Sep 2006 - Dec 2008
drug	N	single b	int c	mut d	single	int	mut	N	ase e	ase
	train	terms	terms		terms	terms		test	Reference	3F
Nucleoside RT inhibitors
AZT	45734	80	108	123	66	77	102	8698	0.091	0.093
3TC	47422	59	64	70	43	52	45	8733	0.059	0.059
ddI	47269	49	21	62	50	25	54	8746	0.054	0.054
d4T	47235	47	34	68	54	20	60	8749	0.050	0.050
ABC	45908	71	46	90	63	24	68	8749	0.048	0.048
FTC	16440	31	35	46	34	34	36	8722	0.086	0.086
TDF	31640	64	91	110	79	83	111	8757	0.065	0.064
Nonnucleoside RT inhibitors
NVP	47400	124	190	142	103	148	110	8729	0.101	0.100
EFV	46054	191	167	211	126	101	142	8687	0.266	0.264
ETR	18166	122	158	160	94	72	119	8493	0.126	0.124

^aJuly-September genotype-phenotype 2006 data was used as validation set for 3F.

^bNumber of single terms (first order effects) in model.

^cNumber of interaction terms in model.

^dNumber of mutations in model.

^eAverage squared error on unseen genotype-phenotype data collected between September 2006 and December 2008.