ncbi-hackathons/determination-of-all-control-datasets-in-large-public-repositories

Putative GEO DataSet prediction

Murat Ozturk, Shuhei Sugai, Sergiusz Wesolowski, Luiz Irber

---

The original task?

Determination of all 'control' samples in public repositories of genomic data.

---

How we abandoned SRA and moved to GEO

SRA Issues:

1. No clear SRA metadata format guidelines.
2. Notoriously non-standard metadata as a result.
3. Control-informative fields are especially sparse (treatment, disease, tumor, affection_status)

---

Switching to a new task on GEO

Automate process of aggregating samples into comparable DataSets.

A DataSet represents a curated collection of biologically and statistically comparable GEO Samples and forms the basis of GEO's suite of data display and analysis tools.

---

Motivation

DataSet curation is manual and backlogged.

NGS experiments should be comparable

• Platforms are machine specific, results should not be machine specific
• We group by Series_type, which is a categorical variable
  • Focus on type "expression profiling by high throughput sequencing"

---

Data source

• Apis mellifera"[porgn:__txid7460]"
• Expression profiling by high throughput sequencing
• 735 Samples (70 Series)

---

Automatic curation: Samples -> DataSets

• Same Platform
• Same organism
• Same experiment type
• Same calibration (hopefully)

---

Practical problems

Even though the search specified full species name, the samples that we have found span 9 different tax-ids

Dead End (again):

• none of the Apis mellifera samples had structured expression data.
• It turns out, GEO does not have a standard for submitting high throughput sequencing data.
• Raw data is deposited to SRA and any processed data is included in 'supplementary data' in free format.

---

We seem to have indepently discovered what is in the GEO FAQ:

Processed sequence data files: GEO hosts processed sequence data files, which are linked at the bottom of Sample and/or Series records as supplementary files. Requirements for processed data files are not yet fully standardized and will depend on the nature of the study, but data typically include genome tracks or expression counts.

---

Not many genes in common

 ('CM000063FS008065472', 3),
 ('CH877218FS000002636', 3),
 ('CM000059FS001816901', 3),
 ('CM000063FS006342682', 3),

---

Conclusion

Our struggles indicate that the problem exists and current data repositories have strucutre far from perfect. To enable the full potential of such resources there must be a reliable way of searching and grouping the samples based on the metadata.

1. Enforce more strict rules on sample submission format with reliable metadata
2. Build a metadata search engine capable of identifying potential mislabling and inconsistencies in the submissions
3. Finally build a sample assembler that will allow the used to find "closest" match control samples for his/her experiment.