Genome Modeling Concepts
이 wiki의 genome modeling concept는 sequence-level과 variant-level AI로 제한합니다. Broad omics나 transcriptomics 섹션이 아니라 Computational Biology의 boundary topic으로 둡니다.
기본 input은 genomic sequence 또는 region입니다.
The model may predict a representation, annotation, variant effect, or generated sequence:
where is optional context such as species, genome assembly, region type, or annotation source.
Route Map
| Need | Start | Risk |
|---|---|---|
| define the object | Genome, Genomic region | reference assembly and coordinate-system mismatch |
| tokenize sequence | K-mer | window overlap and loss of long-range context |
| predict variant impact | Variant effect prediction | assay/source-dependent labels |
| use annotations | Genome annotation | annotation version and target leakage |
| connect to learning | Self-supervised learning, Sequence | pretraining loss may not support downstream claim |
Core Concepts
| Group | Notes |
|---|---|
| Objects | Genome, Sequence, Genomic region |
| Representation | K-mer, Sequence modality |
| Tasks | Variant effect prediction, Genome annotation |
Checks
- What genome assembly and coordinate system are used?
- What is one example: base, k-mer, region, gene, variant, or chromosome segment?
- Are homologous, duplicated, or overlapping regions separated across splits?
- Are labels public, reproducible, and tied to their annotation source?
- Is the task sequence-level, region-level, variant-level, or annotation-level?
Boundary
이 섹션은 의도적으로 broad cell omics, transcriptomics, clinical genomics로 확장하지 않습니다. 그런 주제는 블로그의 research direction에 직접 필요해질 때만 추가합니다.