# Evidence missing from JSON full data download

**URL:** <https://community.opentargets.org/t/evidence-missing-from-json-full-data-download/1443>\
**Category:** Data Access\
**Created:** [3 June 2024 14:04 UTC](https://community.opentargets.org/t/evidence-missing-from-json-full-data-download/1443 "2024-06-03T14:04:34Z")\
**Posts on this page:** 5\
**Page:** 1

<div class="post-metadata">

**Author:** ![technomage](https://dub1.discourse-cdn.com/flex017/user_avatar/community.opentargets.org/technomage/32/632_2.png) [@technomage](https://community.opentargets.org/u/technomage)\
**Post date:** [3 June 2024 14:04 UTC](https://community.opentargets.org/t/evidence-missing-from-json-full-data-download/1443/1 "2024-06-03T14:04:34Z")

</div>

The evidence for [ENSG00000000003/MONDO\_0004992](https://platform.opentargets.org/evidence/ENSG00000000003/MONDO_0004992) indicates europepmc and expression\_atlas datasourceIds, but only the europepmc evidence is present in the 23.09 and 24.03 JSON downloads (7 lines total).

I’m using the data available at [ftp.ebi.ac.uk/pub/databases/opentargets/platform/24.03/output/etl/json/evidence](http://ftp.ebi.ac.uk/pub/databases/opentargets/platform/24.03/output/etl/json/evidence).

---

<div class="post-metadata">

**Author:** ![dsuveges](https://dub1.discourse-cdn.com/flex017/user_avatar/community.opentargets.org/dsuveges/32/30_2.png) [@dsuveges](https://community.opentargets.org/u/dsuveges)\
**Post date:** [3 June 2024 14:19 UTC](https://community.opentargets.org/t/evidence-missing-from-json-full-data-download/1443/2 "2024-06-03T14:19:11Z")

</div>

Hi Tim,

I might misunderstand the questions, but when I fetch data from the linked URL (`https://ftp.ebi.ac.uk/pub/databases/opentargets/platform/24.03/output/etl/json/evidence/sourceId\=expression_atlas/`), the resulting dataset has 200 JSON partitions with 230,182 evidence strings.

```bash
P="https://ftp.ebi.ac.uk/pub/databases/opentargets/platform/24.03/output/etl/json/evidence/sourceId\=expression_atlas/"

wget --recursive --no-parent -e robots=off --cut-dirs=8 $P

cat ftp.ebi.ac.uk/evidence/sourceId=expression_atlas/*json | wc -l # 230182

```

Let me know if there’s anything I have missed.

---

<div class="post-metadata">

**Author:** ![technomage](https://dub1.discourse-cdn.com/flex017/user_avatar/community.opentargets.org/technomage/32/632_2.png) [@technomage](https://community.opentargets.org/u/technomage)\
**Post date:** [3 June 2024 14:33 UTC](https://community.opentargets.org/t/evidence-missing-from-json-full-data-download/1443/3 "2024-06-03T14:33:01Z")

</div>

The `expression_atlas` data is missing from specific disease/target pairs. Sorry the specific example got obscured in the link.

The evidence link for [ENSG00000000003 =\> MONDO\_0004992](https://platform.opentargets.org/evidence/ENSG00000000003/MONDO_0004992) displays expression\_atlas data which is missing from the download.

There are other pairs which are missing the `expression_atlas` data, but I don’t have a list handy ATM.

---

<div class="post-metadata">

**Author:** ![dsuveges](https://dub1.discourse-cdn.com/flex017/user_avatar/community.opentargets.org/dsuveges/32/30_2.png) [@dsuveges](https://community.opentargets.org/u/dsuveges)\
**Post date:** [3 June 2024 19:46 UTC](https://community.opentargets.org/t/evidence-missing-from-json-full-data-download/1443/4 "2024-06-03T19:46:03Z")

</div>

Ah, OK, this UI vs data question also boils down to the discrepancy between direct and indirect evidence. Direct evidence is when the source explicitly tells target x is linked to breast cancer. However, we can take advantage of the disease ontology and we can infer that target x is also linked to cancer (as being an ancestor of breast cancer), this is what we called indirect evidence.

By default the evidence widgets show indirect evidence. Please go the cited [platform page](https://platform.opentargets.org/evidence/ENSG00000000003/MONDO_0004992), find the expression\_atlas widget, click on API Query button in the top-right corner. This will show you the actual graphql query that fetches the data for the widget. You’ll see the `enableIndirect: true` option. Switch it to `false` and you’ll get only direct evidence which will be consistent with the downloadable data.

This is a recurring issue that causes confusion (see [here](https://community.opentargets.org/t/difference-between-direct-and-indirect-links-in-data-downloads/376/4), [here](https://community.opentargets.org/t/how-does-the-platform-display-direct-and-indirect-evidence/341/7) and [here](https://community.opentargets.org/t/spurious-indirect-association-evidence-via-graphql-api/879/2)), we are currently working on a suitable resolution to ensure the association page and the content of the widget is consistent.

---

<div class="post-metadata">

**Author:** ![technomage](https://dub1.discourse-cdn.com/flex017/user_avatar/community.opentargets.org/technomage/32/632_2.png) [@technomage](https://community.opentargets.org/u/technomage)\
**Post date:** [3 June 2024 21:08 UTC](https://community.opentargets.org/t/evidence-missing-from-json-full-data-download/1443/5 "2024-06-03T21:08:43Z")

</div>

Thank you, Daniel. That makes sense.