Table 5. Individual accuracy percentages per experiment in CDR-L1 and -L3, excluding non-predictable (novel) conformations.

The previously acquired clustering set was used for initial DCP training and canonical templates’ updating. The newly downloaded blind dataset was divided in two subsets: for DCP, subset 1 was used for parameter validation (“validation set”), while subset 2 was used for evaluation (“test set”). Both subsets were used for evaluation of canonical templates, as no parameterisation was necessary, however the terms “validation” and “test” were retained for the two subsets for disambiguation and in order to allow direct comparisons. In post-evaluation Phase-2, the validation set was merged to the clustering set for DCP re-training and canonical templates’ re-updating. Updated methods were then evaluated on the test set that remained blind, but also were applied for retro-prediction on the validation set.

CDR-L1 predictions
Phase-1 Initial DCP signatures		Phase-2 Updated DCP signatures		Phase-1 Initial canonical templates		Phase-2 Updated canonical templates
Training: clustering set Evaluation: validation set	Training: clustering set Evaluation: test set	Training: clustering + validation sets Evaluation: validation set	Training: clustering + validation sets Evaluation: test set	Template Updating: clustering set Evaluation: validation set	Template Updating: clustering set Evaluation: test set	Training: clustering + validation sets Evaluation: validation set	Template Updating: clustering + validation sets Evaluation: test set
99% (86/87)	99% (77/78)	100% (87/87)	98% (76/78)	92% (80/87)	96% (75/78)	98% (85/87)	96% (75/78)
Cumulative evaluation on validation + test sets		Cumulative evaluation on validation + test sets		Cumulative evaluation on validation + test sets		Cumulative evaluation on validation + test sets
99% (163/165)		99% (163/165)		94% (155/165)		97% (160/165)
CDR-L3 predictions
Phase-1 Initial DCP signatures		Phase-2 Updated DCP signatures		Phase-1 Initial canonical templates		Phase-2 Updated canonical templates
Training: clustering set Evaluation: validation set	Training: clustering set Evaluation: test set	Training: clustering + validation sets Evaluation: validation set	Training: clustering + validation sets Evaluation: test set	Template Updating: clustering set Evaluation: validation set	Template Updating: clustering set Evaluation: test set	Training: clustering + validation sets Evaluation: validation set	Template Updating: clustering + validation sets Evaluation: test set
95% (84/88)	89% (70/79)	100% (88/88)	91% (72/79)	95% (69/73)	87% (62/71)	100% (73/73)	89% (63/71)
Cumulative evaluation on validation + test sets		Cumulative evaluation on validation + test sets		Cumulative evaluation on validation + test sets		Cumulative evaluation on validation + test sets
92% (154/167)		96% (160/167)		91% (131/144)		94% (136/144)