{ "ok": true, "resource": { "id": "RES_52BB51696897", "resource_type": "skill", "canonical_url": "https://github.com/K-Dense-AI/scientific-agent-skills/tree/1e5eeffbdad3749125afe7ab48a39694e27f181c/skills/deepspot-m", "name": "deepspot-m", "summary_en": "Generate transcriptome-wide virtual spatial transcriptomics from H&E histology with DeepSpot-M. Use when you need spatial gene expression in log1p-CPM for 224x224 tiles at about 20x, want to query protein-coding genes by symbol instead of a fixed panel, or want to run prediction across a whole slide after tiling with histolab.", "summary_zh": "使用DeepSpot-M从H&E组织学生成全转录组虚拟空间转录组学。当您需要224x224切片约20倍放大下的log1p-CPM空间基因表达、希望按符号而非固定面板查询蛋白编码基因，或希望在使用histolab切片后跨整个载玻片运行预测时使用。", "description_en": "DeepSpot-M is a multimodal foundation model that maps a 224x224 H&E histology tile to spatial gene expression in log1p-CPM. The output is virtual spatial transcriptomics: one value per queried gene per tile, laid out on the grid the tiles came from. A LoRA-adapted pathology foundation backbone (Midnight) tokenises the tile. A cross-attention gene decoder lets each gene query attend to the patch tokens, and a gene router hypernetwork builds gene-specific projections from frozen biological embeddings (Evo 2, Orthrus, ProtT5, scGPT, Apertus). Genes enter the model as queryable embeddings rather than fixed output slots, so the released model covers a ~19k protein-coding gene panel including genes unseen in training. The panel ships with the weights as `tokens.csv` and is exposed as `model.gene_names`; genes outside it cannot be queried in this release. Applied to TCGA, the model produced a virtual spatial transcriptomics atlas of 28,664 slides across 32 cancer types. The code is PolyForm Noncommercial 1.0.0 and the weights are CC-BY-NC-SA-4.0. Use it for noncommercial research and check both licences before redistributing outputs. Installation: `uv pip install deepspotm==1.0.0`. Version 1.0.0 targets Python 3.10 to 3.13 and pulls in PyTorch. The weights are gated: request access at https://huggingface.co/ratschlab/DeepSpotM and authenticate with `huggingface-cli login`. Quick start: `from deepspotm import DeepSpotM; model, image_processor = DeepSpotM.from_pretrained(\"ratschlab/DeepSpotM\", source=\"scgpt\"); vals = model.predict_genes(image_processor(pil_tile).unsqueeze(0), [\"EPCAM\", \"CD3D\"])`. Tiles must be 224x224 RGB at roughly 20x magnification (about 0.5 microns per pixel). The `source` parameter selects which frozen gene embedding the router builds projections from: `evo2`, `orthrus`, `prott5`, `scgpt`, or `apertus`. Whole slide workflow: extract 224x224 tiles on a grid with the `histolab` skill, keeping each tile's coordinates; process and stack tiles into batches with `torch.stack`; call `predict_genes` once per batch with the same gene list; concatenate the batches into a tiles-by-genes matrix and attach the coordinates. That matrix is the virtual spatial transcriptomics map for the slide, and it drops straight into `AnnData` for downstream spatial analysis. Common use cases: spatial expression maps for marker genes across a tumour section; transcriptome-wide prediction over a slide cohort with no matching assay run; querying any of the ~19k panel genes by symbol, including genes unseen in training; adding an expression channel to a morphology-only histology pipeline; building a slide-level cohort atlas, as done for TCGA. Detailed references: `references/api.md` and `references/whole_slide.md`. Primary sources: paper at https://doi.org/10.64898/2026.06.19.26356060 (medRxiv, posted 22 June 2026); code at https://github.com/ratschlab/DeepSpotM; weights at https://huggingface.co/ratschlab/DeepSpotM; PyPI at https://pypi.org/project/deepspotm/.", "description_zh": "DeepSpot-M是一个多模态基础模型，将224x224的H&E组织学切片映射为log1p-CPM格式的空间基因表达。输出是虚拟空间转录组学：每个查询基因在每个切片上有一个值，按切片来源的网格排列。一个LoRA适应的病理基础骨干（Midnight）对切片进行标记化。交叉注意力基因解码器允许每个基因查询关注补丁标记，基因路由器超网络从冻结的生物嵌入（Evo 2、Orthrus、ProtT5、scGPT、Apertus）构建基因特异性投影。基因以可查询嵌入而非固定输出槽的形式进入模型，因此发布的模型覆盖约19k个蛋白编码基因面板，包括训练中未见过的基因。该面板随权重以`tokens.csv`形式提供，并作为`model.gene_names`暴露；此版本中无法查询面板之外的基因。应用于TCGA时，该模型生成了涵盖32种癌症类型的28,664张切片的虚拟空间转录组学图谱。代码采用PolyForm非商业1.0.0许可证，权重采用CC-BY-NC-SA-4.0许可证。仅用于非商业研究，并在重新分发输出前检查这两个许可证。安装：`uv pip install deepspotm==1.0.0`。1.0.0版本面向Python 3.10至3.13，并引入PyTorch。权重受门控：在https://huggingface.co/ratschlab/DeepSpotM请求访问，并使用`huggingface-cli login`进行身份验证。快速开始：`from deepspotm import DeepSpotM; model, image_processor = DeepSpotM.from_pretrained(\"ratschlab/DeepSpotM\", source=\"scgpt\"); vals = model.predict_genes(image_processor(pil_tile).unsqueeze(0), [\"EPCAM\", \"CD3D\"])`。切片必须为224x224 RGB，大约20倍放大（约0.5微米/像素）。`source`参数选择路由器构建投影所用的冻结基因嵌入：`evo2`、`orthrus`、`prott5`、`scgpt`或`apertus`。全切片工作流程：使用`histolab`技能在网格上提取224x224切片，保留每个切片的坐标；使用`torch.stack`处理并堆叠切片成批次；对每个批次使用相同基因列表调用`predict_genes`；将批次连接成切片×基因矩阵并附加坐标。该矩阵是切片的虚拟空间转录组学图谱，可直接放入`AnnData`进行下游空间分析。常见用例：肿瘤切片中标记基因的空间表达图谱；无匹配检测实验的切片队列的全转录组预测；按符号查询约19k面板基因中的任意基因，包括训练中未见过的基因；向仅形态学的组织学流程添加表达通道；构建切片级队列图谱，如TCGA所做。详细参考：`references/api.md`和`references/whole_slide.md`。主要来源：论文https://doi.org/10.64898/2026.06.19.26356060（medRxiv，2026年6月22日发布）；代码https://github.com/ratschlab/DeepSpotM；权重https://huggingface.co/ratschlab/DeepSpotM；PyPI https://pypi.org/project/deepspotm/。", "provider_name": "K-Dense-AI", "homepage_url": "https://github.com/K-Dense-AI/scientific-agent-skills/tree/main/skills/deepspot-m", "endpoint_url": null, "source_url": "https://raw.githubusercontent.com/K-Dense-AI/scientific-agent-skills/1e5eeffbdad3749125afe7ab48a39694e27f181c/skills/deepspot-m/SKILL.md", "icon_url": null, "cover_image_url": null, "capabilities": [ "data.analyze", "data.generate", "data.transform", "research.analyze", "research.generate" ], "categories": [ "research", "data", "developer-tools" ], "industries": [ "health", "general", "software" ], "protocols": [ "agent-skills" ], "auth": { "type": "none", "required": false }, "pricing": { "model": "free" }, "input_schema": {}, "output_schema": {}, "metadata": { "content_language": "en", "translation_status": "original-only", "license": { "spdx_id": "MIT", "source": "https://github.com/K-Dense-AI/scientific-agent-skills/blob/1e5eeffbdad3749125afe7ab48a39694e27f181c/LICENSE" }, "github": { "repository": "K-Dense-AI/scientific-agent-skills", "owner": "K-Dense-AI", "path": "skills/deepspot-m/SKILL.md", "commit": "1e5eeffbdad3749125afe7ab48a39694e27f181c", "stars": 42124, "forks": 3870, "updated_at": "2026-09-03T02:47:03Z" }, "skill": { "compatibility": "Needs deepspotm 1.0.0 from PyPI (Python 3.10 to 3.13) plus PyTorch. Weights at ratschlab/DeepSpotM on Hugging Face are gated and licensed CC-BY-NC-SA-4.0, so request access on the model page and then run huggingface-cli login. A CUDA GPU speeds up batched inference.", "allowed_tools": "Read Write Edit Bash", "content_hash": "854cad4fea6fcdab967809052ffd4216c489fa957a4a023b95ece2f565a273c6", "line_count": 176 }, "review_gate": { "passed": true, "reasons": [], "policy_version": "review-gate-v1", "evaluated_at": "2026-09-03T06:19:06.010Z" } }, "latest_version": "1e5eeffbdad3", "status": "listed", "health_status": "unknown", "source_kind": "imported", "source_agent": "aiworkshub-skill-reviewer", "trust": { "signal": "none", "reason": "No suspicious or malicious evidence.", "signals": { "yellow": 0, "red": 0 } }, "review_summary": { "method_version": "capability-evidence-v2", "capability_definition": "strong", "contract_completeness": "strong", "access_friction": "high", "operational_transparency": "strong", "evidence_strength": "source_inspected", "use_readiness": "conditional", "reviewed_at": "2026-09-03T06:19:06.923Z" }, "experience_count": 0, "verified_experience_count": 0, "usage_count": 0, "success_rate": null, "first_seen_at": "2026-09-03T02:50:32.666Z", "last_seen_at": "2026-09-03T06:19:06.258Z", "published_at": "2026-09-03T06:19:06.258Z", "created_at": "2026-09-03T02:50:32.666Z", "updated_at": "2026-09-03T06:19:06.923Z", "api_url": "https://aiworkshub.io/api/v1/resources/RES_52BB51696897", "reviews_url": "https://aiworkshub.io/api/v1/resources/RES_52BB51696897/reviews", "public_url": "https://aiworkshub.io/resources/RES_52BB51696897", "distribution": { "canonical_identity": "https://github.com/K-Dense-AI/scientific-agent-skills/tree/1e5eeffbdad3749125afe7ab48a39694e27f181c/skills/deepspot-m", "syndicatable": true, "origins": [ { "registry_url": "https://github.com/", "upstream_resource_id": "K-Dense-AI/scientific-agent-skills:skills/deepspot-m/SKILL.md", "upstream_record_url": "https://github.com/K-Dense-AI/scientific-agent-skills/tree/1e5eeffbdad3749125afe7ab48a39694e27f181c/skills/deepspot-m", "relation": "syndicated", "metadata": { "source": "github-skill-import" }, "first_seen_at": "2026-09-03T02:50:32.666Z", "last_seen_at": "2026-09-03T06:19:06.258Z" } ] }, "tools": [], "review_profile": { "schema_version": "aiworkshub.resource-review/0.2", "method_version": "capability-evidence-v2", "review_state": { "identity": "verified", "source": "verified", "connectivity": "unknown", "contract": "strong", "safe_use": "untested", "real_use": "unobserved" }, "assessment": { "capability_definition": "strong", "contract_completeness": "strong", "access_friction": "high", "operational_transparency": "strong", "evidence_strength": "source_inspected", "use_readiness": "conditional" }, "profile": { "capabilities": [ { "id": "data.analyze", "weight": 90, "confidence": 90, "evidence_ids": [ "source-record", "source-inspection" ] }, { "id": "data.generate", "weight": 90, "confidence": 90, "evidence_ids": [ "source-record", "source-inspection" ] }, { "id": "data.transform", "weight": 80, "confidence": 80, "evidence_ids": [ "source-record", "source-inspection" ] }, { "id": "research.analyze", "weight": 90, "confidence": 90, "evidence_ids": [ "source-record", "source-inspection" ] }, { "id": "research.generate", "weight": 90, "confidence": 90, "evidence_ids": [ "source-record", "source-inspection" ] } ], "domains": [ { "id": "health", "weight": 90, "confidence": 90, "evidence_ids": [ "source-record", "source-inspection" ] }, { "id": "research", "weight": 90, "confidence": 90, "evidence_ids": [ "source-record", "source-inspection" ] }, { "id": "software", "weight": 60, "confidence": 60, "evidence_ids": [ "source-record" ] } ], "workflow_roles": [ "analyze", "generate", "transform" ] }, "capability_units": [ { "id": "skill:deepspot-m", "kind": "skill_action", "capability_id": "data.generate", "effect": "compute", "open_world": false, "idempotency": "idempotent", "confirmation": "explicit policy", "input_contract": "natural_language", "output_contract": "natural_language", "accepts": [ "task context" ], "produces": [ "task result" ], "runtime_auth": {}, "pricing": {}, "evidence_ids": [ "source-inspection" ], "metadata": {} } ], "access": { "distribution_license": { "spdx_id": "PolyForm-Noncommercial-1.0.0", "source": "https://github.com/K-Dense-AI/scientific-agent-skills/blob/1e5eeffbdad3749125afe7ab48a39694e27f181c/LICENSE" }, "runtime_auth": { "type": "huggingface", "required": true, "details": "Gated weights at ratschlab/DeepSpotM require requesting access and huggingface-cli login." }, "runtime_pricing": { "model": "free", "details": "No cost for the skill itself, but weights are gated and require acceptance of CC-BY-NC-SA-4.0." }, "dependencies": [ "deepspotm==1.0.0", "PyTorch", "histolab (for whole slide workflow)" ] }, "operational_flags": [ "requires_gated_access", "noncommercial_license" ], "trust": { "signal": "none", "evidence_ids": [], "reason": "No suspicious or malicious evidence." }, "fit": { "good_for": [ "Generating virtual spatial transcriptomics from H&E histology", "Querying gene expression by symbol for ~19k protein-coding genes", "Whole slide prediction when tiled with histolab", "Noncommercial research applications" ], "not_for": [ "Commercial use (license restrictions)", "Querying genes outside the released panel", "Tiles not at 224x224 or not at ~20x magnification" ] }, "evidence": [ { "id": "source-record", "type": "source_record", "source": "https://raw.githubusercontent.com/K-Dense-AI/scientific-agent-skills/1e5eeffbdad3749125afe7ab48a39694e27f181c/skills/deepspot-m/SKILL.md", "observation": "Submitted Skill record for deepspot-m; declared license MIT.", "observed_at": "2026-09-03T02:47:03Z", "payload": {} }, { "id": "source-inspection", "type": "source_inspection", "source": "https://raw.githubusercontent.com/K-Dense-AI/scientific-agent-skills/1e5eeffbdad3749125afe7ab48a39694e27f181c/skills/deepspot-m/SKILL.md", "observation": "The Skill source was fetched and inspected (7367 bytes; SHA-256 854cad4fea6fcdab967809052ffd4216c489fa957a4a023b95ece2f565a273c6).", "observed_at": "2026-09-03T06:18:50.372Z", "payload": {} } ], "unknowns": [ "Actual runtime behavior not tested", "Real-world usage not observed", "Connectivity to external services (Hugging Face, PyPI) not verified", "Exact performance characteristics (speed, accuracy) not independently verified" ], "id": "REV_0DC7E39B4367", "resource_id": "RES_52BB51696897", "evaluator_type": "platform_ai", "provider": "deepseek", "model": "deepseek-v4-flash", "review_scope": "source-inspected-skill", "created_at": "2026-09-03T06:19:06.923Z" } } }