# FAQ

Below are answers to the most commonly asked questions about K Pro. If you don't find your answer here, please contact us via the [support form](https://owkinkhelp.zendesk.com/hc/en-us/requests/new?tf_anonymous_requester_email=ewa.kondratowicz-ext@owkin.com\&tf_34057044510353=https://k.owkin.com/chat).

### Getting Started

**What is K Pro and how is it different from a general-purpose LLM?** K Pro is Owkin's agentic AI platform purpose-built for biomedical research. Unlike generic LLMs, K Pro (1) grounds every answer in queried patient data and validated scientific sources rather than parametric memory alone, (2) provides access to exclusive multimodal datasets such as MOSAIC, (3) generates interactive, publication-ready visualizations in real time, (4) coordinates specialized AI agents — each designed for a distinct research task — through an orchestration layer, and (5) is engineered by scientists for biomedical workflows.

**What are the available service tiers?** K Pro is available in four tiers — Free, Light, Standard, and Premium — each unlocking additional agents, datasets, and collaboration features. K Pro Free gives individual access to the Analyze Agentic Space and public datasets (MOSAIC Window, TCGA). Light adds custom data upload (BYOD) and CCPA compliance. Standard introduces team collaboration and the Activate space. Premium unlocks the Amplify space, dedicated onboarding, SSO, RBAC, and priority support. See the [Service tiers](https://docs.owkin.com/getting-started/service-tiers) page for a full comparison.

**What are Agentic Spaces and K Agents?** Agentic Spaces are thematic groupings of AI agents aligned to phases of the research and drug development pipeline. The three spaces are **Analyze** (literature review, multimodal exploration, gene knowledge), **Activate** (biomarker identification, patient stratification), and **Amplify** (digital twin enrichment, clinical trial strategy). Each space contains specialized K Agents — for example, the Literature Navigator, Multimodal Explorer, and Knowledge Explorer in Analyze. Agents are orchestrated automatically based on the user's query.

**How do I get the best results from K Pro?** Formulate specific, precise queries rather than broad questions. Specify the data types you want to explore (genomic, clinical, spatial, etc.), request visualizations for complex relationships, and use follow-up suggestions to deepen your analysis. Progressing from exploratory to detailed questions yields the best outcomes.

### Data & Datasets

**What datasets are available in K Pro?** All tiers include access to TCGA (20,000+ samples across 33 cancer types) and the MOSAIC Window (60 patients across 5 cancer types). Paid tiers can access the full MOSAIC dataset (2,200+ patients, 9 cancer types, 5 modalities) and additional datasets from the data catalog. From the Light tier onward, users can also upload their own proprietary data (BYOD).

**What data modalities and file formats does K Pro support?** K Pro supports clinical data (`.csv`, `.tsv`, `.xlsx`), bulk RNA-seq (`.txt`, `.tsv`, `.csv`, `.h5ad`), single-cell/nuclei RNA-seq (`.mtx`, `.h5`, `.h5ad`, `.rds`), spatial transcriptomics (`.mtx`, `.h5`, `.h5ad`, `.rds`), whole exome/genome sequencing (`.vcf`), proteomics (`.txt`, `.tsv`, `.csv`, `.h5ad`), H\&E histology (`.tif`, `.tiff`, `.svs`, `.dcm`, `.ndpi`, `.mrxs`), and immunohistochemistry (same imaging formats). See the [Supported modalities](https://docs.owkin.com/data-in-k-pro/k-pro-data-model-and-technical-references/supported-modalities) page for full details.

**How do I upload and prepare my own data?** From the Light tier onward, you can upload proprietary datasets through the Bring Your Own Data (BYOD) capability. Data preparation can be handled in two ways: (1) the **Data Transformation Agent (DTA)**, which provides automated validation and standardization for supported formats, or (2) **expert curation services** for complex datasets, custom experimental formats, or modalities not yet supported by the DTA (e.g., certain imaging data or spatial transcriptomics). See [Preparing your data](https://docs.owkin.com/data-in-k-pro/integrating-your-data-to-k-pro/preparing-your-data) for guidance.

**Can I connect K Pro to my own cloud storage or data platform?** Yes. K Pro supports integration with customer-managed storage solutions, including dedicated cloud accounts and buckets (e.g., AWS), as well as enterprise data platforms such as Databricks and Snowflake. These connections use secure cross-account access patterns to maintain data isolation and governance. See [Connecting your data sources](https://docs.owkin.com/data-in-k-pro/integrating-your-data-to-k-pro/connecting-your-data-sources) for details.

**What is the AI-Readiness Maturity Model?** Owkin's 6-level framework (Level 0–5) for assessing how well a dataset is prepared for use with K Pro. It ranges from uncontrolled data with no governance (Level 0) to fully traceable, AI/ML-optimized datasets with reproducibility standards (Level 5). The model helps organizations understand what steps are needed to make their data AI-ready. See [The AI maturity model](https://docs.owkin.com/data-in-k-pro/k-pro-data-model-and-technical-references/the-ai-maturity-model) for the full framework.

**How does K Pro enrich my data?** Through its data enrichment pipeline, K Pro can augment datasets with AI-derived features including spatial gene expression predictions from H\&E images, cell-type deconvolution at near single-cell resolution, nuclear morphology analysis (cell counts, densities, shape metrics), and ligand-receptor interaction modeling for cell communication analysis. See [Data enrichment](https://docs.owkin.com/data-in-k-pro/integrating-your-data-to-k-pro/data-enrichment) for details.

### Pathology Explorer

**What is Pathology Explorer?** Pathology Explorer is an AI-powered tool that transforms H\&E whole-slide images into granular, queryable insights. Trained on over 200,000 annotations, it detects and classifies 6 cell types (lymphocytes, neutrophils, eosinophils, plasmocytes, fibroblasts, and cancer cells) and produces quantitative features including cell counts, densities, nuclear morphology, spatial co-occurrence, and TIL diffusivity. It currently supports 27 TCGA tumor cohorts.

**How do I access Pathology Explorer through Claude?** Pathology Explorer is available as an MCP (Model Context Protocol) integration. To connect it, add Owkin's MCP server (`https://mcp.k.owkin.com/mcp`) as a Custom Connector in Claude.ai or Claude Desktop (requires a paid Claude plan). You will be prompted to authenticate with your Owkin account via OAuth. Once connected, you can invoke Pathology Explorer tools directly from your Claude interface. See the [Pathology Explorer getting started](https://docs.owkin.com/core-features-and-usage/pathology-explorer-mcp-ai-powered-tissue-analysis/getting-started) guide for step-by-step instructions.

**Can I export data from Pathology Explorer?** Yes. Pathology Explorer supports data export in Parquet format, and slide images can be downloaded through presigned URLs.

### Visualizations

**What types of visualizations does K Pro support?** K Pro supports over 17 chart types spanning clinical data (Kaplan-Meier survival plots, Gantt treatment timelines, Sankey treatment flows), bulk RNA-seq (violin plots, heatmaps, UMAP, pairwise correlation, differential expression), single-cell RNA-seq (cell-type proportion, dot plots, co-expression, UMAP), spatial transcriptomics (slide displays, Moran's I, cell-type co-occurrence, deconvolution bar charts), histomics (cell-type proportion), and cross-modal concordance plots. See [Visualisation capabilities](https://docs.owkin.com/core-features-and-usage/overview/visualisation-capabilities) for a full catalog.

**Can I export plots and analyses from K Pro?** Yes. K Pro generates interactive Plotly visualizations that can be reviewed and exported. The underlying code used to generate each plot is also available for inspection and reproducibility.

### Data Confidentiality & Protection

**How does K Pro protect confidential data?** Only chat history is saved, and it is linked to authenticated users within their organization. The database is a managed RDS instance on Owkin's AWS account, accessible solely through secure service credentials and network policies. Data is stored in an efficient, column-oriented file format optimized for secure storage and retrieval.

**How does K Pro guard against IP leakage?** Data is segregated by customer. Employee access is restricted to those with operational maintenance roles. Security includes 24/7 monitoring across multiple protective layers.

**Can anyone see the data uploaded to K Pro?** No. Uploaded data is visible only to you and your organization members — never shared externally.

**Where is my data hosted?** Data is stored on Owkin's managed, secure cloud infrastructure. For customers with their own cloud subscriptions, data can reside in a separate cloud account maintained distinct from K Pro's infrastructure, with access enabled through secure cross-account access patterns. See [Infrastructure and hosting](https://docs.owkin.com/data-in-k-pro/privacy-and-compliance/infrastructure-and-hosting) for more details.

### Transparency & Explainability

**Can I access the codebase and decision-making process of K Pro?** While K Pro's codebase is proprietary, users can inspect the reasoning process undertaken by agents. Generated plots and the code behind them are available for review, providing full transparency into how results are produced.

**How can I make sure that the scientific conclusions generated by K Pro are evidence-based and accountable?** Results are grounded in authoritative sources like PubMed and validated biological knowledge bases. K Pro provides complete provenance for all outputs — users can trace recommendations back to original data sources and citations through explainability features that log all data sources, model decisions, and reasoning steps.

**How does K Pro handle data integration and standardization across diverse datasets?** K Pro uses specialized tools and agents designed for dataset diversity. The orchestration layer captures user intent and calls appropriate tools to query scientific literature, gene databases, and multi-omics cohorts. Under the hood, data is harmonized following an OMOP-like schema and preferred ontologies to ensure consistent querying across heterogeneous sources.

### Citations & References

**How does Owkin ensure proper and deterministic retrieval of citations and references?** A RAG (Retrieval-Augmented Generation) system verifies the existence and relevance of cited PubMed articles. K Pro's Literature Navigator queries over 22 million PubMed abstracts through semantic search, and internal evaluation against public benchmarks monitors alignment between generated answers and cited sources.

### Trust & Scientific Rigor

**How does Owkin measure and prevent hallucinations?** K Pro employs multiple layers of protection: (1) daily monitoring of Tool Call Accuracy (TCA) — the percentage of interactions where the correct tool is identified and called with accurate parameters, (2) RAG-based citation validation ensuring all referenced PubMed IDs correspond to real articles, (3) tool-based grounding that anchors analysis in modality-specific AI models and actual patient data rather than LLM parametric memory, and (4) automated evaluation tests using curated questions run on a regular basis.

**How does Owkin prevent bias in K Pro recommendations?** Bias audits occur during development and testing phases. Models are evaluated across diverse demographic and clinical datasets with independent validation from academic and clinical partners. Post-deployment continuous monitoring identifies emerging sources of bias. Transparency features enable users to trace data sources and recommendation rationale.

**Who oversees K Pro's AI ethics?** Owkin engages in active collaboration with external stakeholders across regulatory, academic, and clinical domains, including regulatory bodies and bioethics groups. The platform is built with a "biologists building for biologists" philosophy, involving researchers as beta testers and early users to ensure practical alignment with scientific workflows and ethical standards.

### LLM & Technology

**How would different LLMs or versions of the same LLM affect final outcomes?** Outcomes depend on the LLM's capability to understand questions and call appropriate tools. Newer LLM versions typically perform better. Switching between LLMs requires adjustments to prompting and context engineering. K Pro's agentic architecture means that the specialized tools and data pipelines remain constant regardless of the underlying LLM — the model orchestrates, but the scientific analysis is performed by dedicated tools and models.

**Does Owkin own its full end-to-end technology stack? If not, how does Owkin vet its vendors?** Much of the technology is developed in-house, including Owkin's proprietary foundation models (iBOT, H0, H0-mini) and the HIPE model for cell segmentation. Select vendor partners undergo rigorous security assessments and contractual alignment. Owkin maintains ISO 27001:2022 (information security) and ISO 13485:2016 (medical device quality) certifications, ensuring GDPR and HIPAA compliance.

### Compliance & Certifications

**What security certifications does Owkin hold?** Owkin is certified to ISO 27001:2022 (information security management) and ISO 13485:2016 (medical device quality management). The platform is designed for full compliance with GDPR (EU/UK), HIPAA (US), and CCPA (California, from Light tier onward).

**Can K Pro be deployed in my own infrastructure?** K Pro supports flexible deployment models. For customers with their own cloud environments, data can reside in a dedicated cloud account separate from K Pro's infrastructure, with secure cross-account access. The implementation lifecycle includes a specification phase to define deployment architecture and security protocols tailored to your requirements. See the [Implementation lifecycle](https://docs.owkin.com/terms-and-support/support/implementation-lifecycle-and-communication-protocol) page for the full process.

### Ownership & Legal

**Who owns the output from user prompts in K Pro?** Owkin does not claim rights over user-submitted content or outputs. Owkin retains a license to use the content and output to improve its products and services, while protecting academic freedom. See the [Terms and conditions](https://docs.owkin.com/terms-and-support/legal/terms-and-conditions) for the complete legal framework.

### Support & Account

**How do I get support?** Submit requests via the [support form](https://docs.owkin.com/terms-and-support/support/support-and-sla). Support operates during standard business hours (9:00 AM – 6:00 PM Paris Time). Response times depend on priority: P1 (critical, full outage) — 1-hour response, 4-hour resolution; P2 (major issue, no workaround) — 2-hour response, 8-hour resolution; P3 (workaround available) — 2-hour response, 16-hour resolution; P4 (general inquiries) — 2-hour response, 75-hour resolution.

**How do I delete my account?** To permanently delete your K Pro account, contact the Customer Success Team by submitting a request through the [support portal](https://docs.owkin.com/terms-and-support/support/support-and-sla). The team will process your request in a timely manner. Note that certain information may be retained when necessary to meet legal requirements or legitimate operational needs, in compliance with GDPR. See [Account management](https://docs.owkin.com/terms-and-support/support/account-management) for details.