Table 1.

ChromTransfer allows fine-tuning on small input data. Mean overall and per-class (positive: open chromatin, negative: closed chromatin) F1 scores on test set data for HepG2 models fine-tuned after bootstrapping training data (1–75% of original training data, 10 bootstraps). The overall F1 score of the original HepG2 fine-tuned model (100%) is included for reference. Bootstrap sample sizes refers to the total number of training examples, including both positive and negative examples. Standard deviations (sd) across bootstrap estimates are included

Sample (%)	Sample (size of training)	Mean F1 negative	sd F1 negative	Mean F1 positive	sd F1 positive	Mean F1 overall	sd F1 overall
1%	747	0.65	0.22	0.68	0.03	0.67	0.12
5%	3733	0.76	0.01	0.72	0.01	0.74	0.01
10%	7467	0.77	0.01	0.71	0.04	0.74	0.02
15%	11201	0.76	0.02	0.73	0.01	0.75	0.01
20%	14935	0.77	0.01	0.74	0.01	0.75	0.01
25%	18668	0.77	0.01	0.74	0.02	0.75	0.01
50%	37366	0.79	0.01	0.76	0.01	0.77	0.01
75%	56005	0.80	0.01	0.77	0.01	0.79	0.01
100%	74673	0.82	-	0.75	-	0.79	-