Docs · EPIC submission format

As challenge solutions, the participants must submit a gzipped plain text file (.txt.gz extension). File format: one value per line, corresponding to the genomic positions at the test contigs in the bed template template.test.bed.gz, see the challenge data description.

Predictions should be strand-specific; thus, there must be two predictions per position, one for each DNA strand.

The submitted file predictions.txt.gz should look like:

0.123
0.40005
0.3
0.3
0.0
1.0
etc

The values must be within the range of [0; 1] and rounded to 5 digits.

The complete and properly formatted example submission (template.txt) filled with random numbers is provided in template.test.txt.gz for each EPIC species.

* Please note that the offline scoring package provides a validator script (see below) to check if your submission conforms to the EPIC format requirements.

You can also use the make_bed script to convert your txt submission to the bed format for convenient debugging:

bash ./make_bed template.test.bed.gz submission.txt.gz output.bed.gz

Technical details

Quick start: it is sufficient to prepare the submissions for the genomic positions specified in the template.test.bed.gz and reformat the result into a one-column plain text file predictions.txt (+ strand, all positions, followed by - strand, all positions), which should then be gzipped prior to uploading to the EPIC scoring system.

Details: we evaluate the predictions against non-repetitive regions of the test contigs (the 'whitelist').

In a resulting predictions file, the positions inside the interval are expected to be ordered by coordinate ascending (even for minus-strand intervals).

For example, a whitelist can be like:

chr1 12 13
chr2 56 58

Then the bed-template, in turn, will be:

chr1 12 13 . 0 +
chr2 56 58 . 0 +
chr1 12 13 . 0 -
chr2 56 58 . 0 -

The predicted scores in the txt.gz file should correspond to positions in the following order:

Note: there are 6 entries as the chr2 whitelisted region is 2bp long.

chr1:12,+
chr2:56,+
chr2:57,+
chr1:12,-
chr2:56,-
chr2:57,-

Thus, the submitted file with predictions (6 lines) must look like:

For convenience, we profile bed files of whitelisted test intervals (standard conventions apply: 0-based, half-open coordinates).

Whitelists are provided as whitelist.{train,test}.bed.gz

and the template is strand-sorted (+ first, - second), see template.{train,test}.bed.gz