2026-08-09 18:32:12 +00:00
1 changed files with 73 additions and 0 deletions
@@ -0,0 +1,73 @@
+---
+title: "AI4Bio Landscape Database"
+task: ""
+lineage_type: import
+upstream_source: https://github.com/inoue0426/awesome-computational-biology/blob/7a064bf0/docs/AI4BIO_LANDSCAPE.md
+upstream_sha: 7a064bf0
+imported_at: 2026-08-08
+prompt_class: catalogue
+upstream_changes: accepted
+author: upstream
+validated: false
+---
+
+# AI4Bio Landscape Database
+
+The landscape view treats the existing computational biology registry as a multidimensional database rather than a single hierarchical list.
+
+## Design goals
+
+- Keep the current curated resource records and generation pipeline intact.
+- Expose orthogonal facets so one resource can be explored by resource type, biological/ML task, data modality, organism, and domain tag.
+- Make the landscape useful without introducing a server or build-time dependency.
+- Keep the data model extensible for richer AI4Bio metadata over time.
+
+## Current facet model
+
+The landscape UI derives the following dimensions from `docs/data/resources.json`:
+
+| Dimension | Source field | Example values |
+|---|---|---|
+| Resource type | `type` | `database`, `benchmark`, `model`, `toolkit`, `api` |
+| Task | `tasks` | `drug-response-prediction`, `cell-type-annotation`, `molecular-generation` |
+| Modality | `modalities` | `transcriptomics`, `spatial-transcriptomics`, `protein-sequence` |
+| Organism | `organism` | `human`, `mouse`, `multi-species` |
+| Domain/tag | `tags` | `drug-discovery`, `single-cell`, `foundation-model` |
+
+These are deliberately treated as separate axes. A model can therefore be, for example, a `model` that performs `perturbation-prediction` on `single-cell-rna-seq` data for `human` and carry tags such as `drug-discovery` and `foundation-model`.
+
+## Recommended schema evolution
+
+The current schema is compatible with a richer landscape database. New fields should be added incrementally and only when they can be curated consistently.
+
+Suggested fields:
+
+| Field | Type | Purpose |
+|---|---|---|
+| `entities` | array of strings | Biological entities such as `gene`, `protein`, `compound`, `cell`, `disease` |
+| `methods` | array of strings | Method families such as `transformer`, `gnn`, `diffusion`, `optimal-transport` |
+| `organizations` | array of strings | Primary organizations responsible for the resource |
+| `year` | integer | Initial public release/publication year |
+| `github` | string | Source repository when distinct from the canonical landing page |
+| `documentation` | string | Documentation URL |
+| `maintenance_status` | string | Curated status such as `active`, `maintenance`, `archived`, `unknown` |
+| `last_checked` | string | Date the metadata/link was last manually or automatically checked |
+
+Avoid adding dynamic popularity metrics such as GitHub stars directly to canonical records unless a reproducible refresh pipeline is introduced. Such values become stale quickly and should be stored as generated metadata rather than curated facts.
+
+## Canonical-source policy
+
+At present, `README.md` is the canonical curated list, with generated YAML/JSON/CSV artifacts. The landscape page intentionally consumes `docs/data/resources.json` without changing that policy.
+
+A future migration may make `data/resources.yml` the canonical source once all README-only categorization semantics can be represented explicitly in structured fields. That migration should be a separate change because it changes contribution workflow and source-of-truth semantics.
+
+## Landscape page
+
+Open `docs/landscape.html` through GitHub Pages. It provides:
+
+- full-text search across names, descriptions, tasks, modalities, organisms, and tags;
+- filters for type, task, modality, organism, and tag;
+- summary counts for resources and major dimensions;
+- frequency bars recalculated for the current filtered result set;
+- direct resource and paper links;
+- client-side rendering with no additional dependencies.