Table 2.

The top 21 6-mers: Z-DNABERT attention rank versus the 6-mer frequency rank in the experimental datasets tested for tuning the model.

	hg38 Kouzine et al		hg38 Shin et al
Attention rank	6-mer	Frequency	6-mer	Frequency
1	GCGCGC	1	TGTGTG	1
2	GTGTGT	5	GTGTGT	2
3	CGCGCG	2	CGCGCG	4
4	ACACAC	6	GCGCGC	3
5	TGTGTG	3	CACACA	5
6	GCGCGG	7	ACACAC	6
7	CACACA	4	GGGGAA	40
8	CCGCGC	10	AAAAAA	17
9	GGGCGC	11	CAGGGA	43
10	GCGCCC	12	GTGCGC	11
11	GTGCGC	17	TGGGGA	331
12	GGCGCG	9	GGGGGA	39
13	GTGTGC	14	GCTGGG	9
14	GCGCAC	19	GTGTGC	7
15	GCACAC	15	TGCGCG	8
16	GCCCGC	20	TGCATG	21
17	GCGGGC	16	GGGAAG	33
18	CGCGCC	8	AGGGAG	429
19	GCGTGC	25	GGGAGC	458
20	GCACGC	26	AGAAAG	38
21	CCCGCG	18	GGGAAA	80

The model based on the experimental Kouzine et al data was used in the paper rather than the much smaller 150 bp resolution ChIP-seq data of Shin et al.