# Wiki Index

Master catalog of all pages. Updated on every ingest.

## Concepts

| Page | One-line summary |
|------|-----------------|
| [Virtual Cell](concepts/virtual-cell.md) | In silico cell model that simulates state and response to perturbations |
| [Perturbation Biology](concepts/perturbation-biology.md) | Systematic study of how cells respond to controlled genetic/chemical/environmental interventions |
| [Single-Cell Foundation Models](concepts/single-cell-foundation-models.md) | Large transformers pretrained on scRNA-seq; debate over whether they surpass simple baselines |

## Papers

| Page | Year | One-line summary |
|------|------|-----------------|
| [scGPT](papers/scgpt.md) | 2024 | Bo's GPT-style model trained on 33M cells; SOTA across 5 single-cell tasks |
| [GEARS](papers/gears.md) | 2023 | Graph + GO prior predicts unseen 2-gene perturbation combinations; Nature Biotechnology |
| [CPA](papers/cpa.md) | 2023 | Compositional autoencoder; disentangles basal state + perturbation + covariate for transfer |
| [scGen](papers/scgen.md) | 2019 | Foundational VAE + vector arithmetic for perturbation transfer across cell types |
| [Norman 2019](papers/norman-2019.md) | 2019 | CRISPRa K562 combinatorial screen; the go-to perturbation benchmark dataset |
| [Replogle 2022](papers/replogle-2022.md) | 2022 | Genome-scale Perturb-seq; 2.5M cells, ~10k gene KOs; largest single-cell perturbation atlas |
| [TranscriptFormer](papers/transcriptformer.md) | 2025 | CZI's 112M-cell cross-species foundation model; spans 12 species, 1.5B years of evolution |
| [Lingshu-Cell](papers/lingshu-cell.md) | 2026 | Discrete diffusion cellular world model; SOTA on Virtual Cell Challenge H1 |
| [Cell-JEPA](papers/cell-jepa.md) | 2026 | JEPA-based pretraining; 36% better clustering than scGPT; representation ≠ perturbation |
| [Scaling Laws for scRNA](papers/scaling-laws-scrna.md) | 2026 | Power-law scaling confirmed for scRNA-seq; data-limited regime = no scaling benefit |
| [Parameter-Free Reps](papers/parameter-free-representations.md) | 2026 | Simple PCA pipelines match foundation models; benchmarks may be too easy |

## Entities

| Page | Type | Description |
|------|------|-------------|
| [Bo Wang](entities/bo-wang.md) | person | Computational biologist; scGPT creator; Xaira Therapeutics |
| [Xaira Therapeutics](entities/xaira-therapeutics.md) | company | Building causal perturbation atlas for Virtual Cell |
| [CellxGene Census](entities/cellxgene.md) | dataset | Largest public scRNA-seq repository; primary FM training corpus |
| [Weissman Lab](entities/weissman-lab.md) | lab | Invented Perturb-seq; generated Norman 2019 + Replogle 2022 datasets |
| [Lotfollahi Lab](entities/lotfollahi-lab.md) | lab | scGen, CPA, chemCPA; foundational perturbation transfer models |

## Questions & Outputs

_(Empty — will populate as Bo asks questions)_

---
*Last updated: 2026-04-04 | Pages: 18 | Sources ingested: 11 papers + seed knowledge*