ChromBERT: Uncovering Chromatin State Motifs in the Human Genome using a BERT-based Approach

Pretrain data and results for: 

Promoter regions for all genes in 127 different cell lines in ROADMAP 

CRM (cis-regulatory module) regions longer than 2k bps in 127 different cell lines in ROADMAP

Whole genome regions without continuous low signal state ("O") in 4-mer

Fine-tuning data and result for:

Promoter regions of high-expressed genes (from RPKM>10 to RPKM>50) compared to not expressed genes (RPKM=0) or low-expressed genes (RPKM>0) with the directory names:

not_n_rpkm0 : RPKM=0 vs. RPKM>0

not_n_rpkm10 : RPKM=0 vs. RPKM>10

not_n_rpkm20 : RPKM=0 vs. RPKM>20 

not_n_rpkm30 : RPKM=0 vs. RPKM>30 

not_n_rpkm50 : RPKM=0 vs. RPKM>50 

rpkm0_n_rpkm10 : RPKM>0 vs. RPKM>10 

rpkm0_n_rpkm20 : RPKM>0 vs. RPKM>20 

rpkm0_n_rpkm30 : RPKM>0 vs. RPKM>30 

rpkm0_n_rpkm50 : RPKM>0 vs. RPKM>50 

rpkm10_n_rpkm20 : RPKM>10 vs. RPKM>20 

rpkm10_n_rpkm30 : RPKM>10 vs. RPKM>30 

rpkm10_n_rpkm50 : RPKM>10 vs. RPKM>50 

rpkm20_n_rpkm30 : RPKM>20 vs. RPKM>30 

rpkm20_n_rpkm50 : RPKM>20 vs. RPKM>50 

rpkm30_n_rpkm50 : RPKM>30 vs. RPKM>50 

CRM regions longer than 2k bps compared to non-CRM regions