Datasets:

yashvir
/

seq2state-data

Does adding the sequence counterfactual feature improve AUPRC over the strongest trained molecular model, on identical matched rows? adds_value is True only when the paired region-bootstrap 95% CI of (blend - trained) is entirely above zero.

panel	trained_model	scoreboard	auprc_trained	auprc_feature	auprc_blend	diff	ci_low	ci_high	adds_value
top250	observed_hic_reabc	strict_all	0.7278	0.6368	0.7107	-0.0171	-0.0374	+0.0025	False
top250	observed_hic_reabc	high_confidence	0.9159	0.8412	0.9015	-0.0145	-0.0345	+0.0035	False
top250	observed_hic_reabc	field_compatible_e2g	0.9380	0.8704	0.9266	-0.0114	-0.0312	+0.0076	False
top250	observed_hic_reabc	very_high_confidence	0.9264	0.8674	0.9138	-0.0126	-0.0374	+0.0109	False
top250	ensemble_proxy_residual_v1	strict_all	0.7314	0.6368	0.7162	-0.0152	-0.0369	+0.0055	False
top250	ensemble_proxy_residual_v1	high_confidence	0.9216	0.8412	0.9069	-0.0147	-0.0353	+0.0037	False
top250	ensemble_proxy_residual_v1	field_compatible_e2g	0.9396	0.8704	0.9301	-0.0095	-0.0296	+0.0096	False
top250	ensemble_proxy_residual_v1	very_high_confidence	0.9398	0.8674	0.9204	-0.0194	-0.0470	+0.0080	False
all_positive	observed_hic_reabc	strict_all	0.6958	0.5885	0.6754	-0.0203	-0.0403	+0.0011	False
all_positive	observed_hic_reabc	high_confidence	0.9044	0.8014	0.8772	-0.0272	-0.0460	-0.0068	False
all_positive	observed_hic_reabc	field_compatible_e2g	0.9229	0.8235	0.8973	-0.0256	-0.0437	-0.0085	False
all_positive	observed_hic_reabc	very_high_confidence	0.8858	0.8186	0.8770	-0.0088	-0.0420	+0.0203	False
all_positive	ensemble_proxy_residual_v1	strict_all	0.6872	0.5885	0.6752	-0.0120	-0.0325	+0.0090	False
all_positive	ensemble_proxy_residual_v1	high_confidence	0.9081	0.8014	0.8807	-0.0274	-0.0459	-0.0081	False
all_positive	ensemble_proxy_residual_v1	field_compatible_e2g	0.9227	0.8235	0.8995	-0.0232	-0.0415	-0.0059	False
all_positive	ensemble_proxy_residual_v1	very_high_confidence	0.9072	0.8186	0.8870	-0.0202	-0.0551	+0.0126	False