Clarification on DNABERT Pre-training Data Format

[vkaz39]

Hello,

I have a few questions regarding your statement:

Please see the template data at /example/sample_data/pre. If you are trying to pre-train DNABERT with your own data, please process your data into the same format as it.

Could you clarify the exact steps you followed to create this pre-training data format? Specifically:

Did you concatenate multi-line FASTA sequences into a single continuous sequence per chromosome, then remove the FASTA headers (sequence IDs)?

After that, did you split each chromosome sequence into 512 bp blocks, with each block written as a separate line in the file? (e.g., block 1 of chromosome 1 = line 1, block 2 of chromosome 1 = line 2, … block 1 of chromosome 4 = line 10, block 2 of chromosome 4 = line 11, etc.)

Was each 512 bp block then k-merized (e.g., with k=6) before being used for training?

Additionally, I noticed your example data does not contain k-mers with “N”. Were those filtered out, or were they simply not present in the input sequences?

Thanks