Maximizing gain in high-throughput screening using conformal prediction

Table 5 Average percent loss in gain where training data did not correctly predict maximum gain for the test set

Cost	Total number of partially screened datasets^a	Fingerprint based models		Physiochemical based models
Cost	Total number of partially screened datasets^a	Number of dataset^b	%loss	Number of dataset^b	%loss
6	9	6	5.7^c	4	2.1
10	10	3	1	3	1.8
14	10	3	1.6	2	0.4

^aDatasets where the validation did not indicate that the entire set should be screened for maximum gain
^bDatasets where the optimum training set validation setting did not correspond to the maximum test set gain
^cFails for dataset 2326: 23.9%. Excluding this result: 2.1%

ISSN: 1758-2946