| --- |
| viewer: false |
| license: cc-by-nc-sa-4.0 |
| task_categories: |
| - tabular-classification |
| - feature-extraction |
| tags: |
| - genomics |
| - bioinformatics |
| - raw-reads |
| - unmapped-sequences |
| - synthetic-biology |
| pretty_name: CGSC Raw Transcriptome Unmapped Reads |
| size_categories: |
| - 100K<n<1M |
| --- |
| |
| # Dataset Card for CGSC Raw Transcriptome Unmapped Reads |
|
|
| ## Dataset Description |
|
|
| - **Homepage:** https://comp-genomics.dev/datasets/transcriptome-raw |
| - **Repository:** Computational Genomics & Synthetic Bio-Arrays Consortium |
| - **Paper:** N/A (Internal Lab Protocol: CGSC-SEQ-2026) |
| - **Point of Contact:** sequence-data@comp-genomics.dev |
|
|
| ### Dataset Summary |
|
|
| This repository acts as the primary cold-storage for unmapped, raw sequencing outputs generated during the Q2 2026 Synthetic Bio-Arrays trials. The dataset comprises massive, uncompressed binary blobs that represent pre-alignment genomic data directly from the sequencing hardware. |
|
|
| Because these files bypass standard alignment and compression algorithms (such as BAM/CRAM conversion) to preserve base-pair quality scores and hardware-level artifact data, the payloads are exceptionally large and entirely unstructured to standard viewers. This dataset is intended exclusively for testing high-throughput bioinformatics ingestion pipelines and error-correction models. |
|
|
| ### Supported Tasks and Leaderboards |
|
|
| - `error-correction-modeling`: Evaluating algorithms that filter sequencing hardware noise from unmapped reads. |
| - `pipeline-stress-testing`: Ingestion of massive, unstructured biological data streams. |
|
|
| ### Languages |
|
|
| There is no natural language present. Metadata and internal logging strings are generated in English (`en`), but the core payload consists of binary-encoded nucleotide sequences and quality scores. |
|
|
| ## Dataset Structure |
|
|
| ### Data Instances |
|
|
| The architecture of this storage does not follow typical huggingface tabular formats. Records point to dense, opaque payloads. |
|
|
| ```json |
| { |
| "run_id": "cgsc-seq-run-8824", |
| "sequencer_node": "illumina-synthetic-node-04", |
| "read_type": "unmapped_paired_end", |
| "compression_state": "raw_binary_dump", |
| "payload_reference": "reads/run-8824-unmapped.bin", |
| "gc_content_estimate": "N/A - Pre-analysis" |
| } |
| |
| ``` |
| ### Data Fields |
| * run_id: Unique batch identifier for the sequencing run. |
| * sequencer_node: Hardware identifier for the machine that generated the data. |
| * read_type: Categorization of the raw output (e.g., unmapped_paired_end). |
| * compression_state: Always raw_binary_dump to justify the immense file weight. |
| * payload_reference: Direct pointer to the large LFS files. |
| * gc_content_estimate: Left as N/A due to the raw nature of the files. |
| ## Dataset Creation |
| ### Curation Rationale |
| Traditional genomic databases host heavily curated and compressed sequences. The CGSC requires a repository of raw, artifact-heavy sequencer dumps to train next-generation base-calling AI models. |
| ### Source Data |
| Data is fully synthetic, generated via advanced transcriptome simulations modeling noisy hardware environments. No real human or animal genetic material is represented in this repository. |
| ## Disclaimers |
| Due to the absence of standard .fastq or .bam formatting, these unstructured binary dumps are massive. Downloads are recommended only for consortium partners with appropriate HPC (High-Performance Computing) storage infrastructure. We provide no parsing tools for external use. |
| ``` |
| |
