How K-Pro works

When you ask K-Pro a question, it routes your request to the right agents and tools, runs the analysis, and returns an interactive result. This page covers the three ways you can work with K-Pro, the agents and tools behind it, and the role of skills.

Three ways to ask K-Pro a question

K-Pro is designed as a no-code platform for wet-lab scientists and disease-area experts. It supports three modes of interaction along a spectrum of control, and a scientist moves between them as the task requires.

Co-pilot: Conversational exploration

Ask. Explore. Iterate.

Ask a question in natural language. Get back an interactive result you can explore and build on.

Example: The scientist opens K-Pro and asks a question in plain English: "Show me NECTIN4 expression across malignant cell types in MOSAIC bladder cancer cohorts, and compare to GTEx healthy tissues." K-Pro returns an interactive cell-type expression plot, a healthy-tissue comparison, and a narrative summary that states which data and statistical tests were used.

The scientist can hover over the plot for details, and ask a plain-English follow-up ("Now show the spatial distribution of NECTIN4 expression relative to CD8+ T cells") to trigger the next analysis.

Semi-autonomous: Skilled workflows

Run expert workflows on demand.

Trigger a skill (Owkin-certified or custom) for complex multi-step analyses where you want a constrained, reproducible workflow rather than free-form conversation.

Example: The scientist knows what analytical workflow they want and invokes a skill directly with a slash command, for example /target-characterization NECTIN4 in BLCA. K-Pro runs the full multi-step workflow end-to-end and returns a structured result. Skills come from three sources:

• Owkin-certified workflows built from a decade of pharma engagements
• KOL-derived skills encoding leading domain experts' methodologies
• Customer-authored skills built by client teams

Autonomous: Discovery campaigns

Scale across hypotheses.

Launch multiple hypotheses in parallel through dedicated sandboxed agents, for example to evaluate several strategies side by side.

Example: The scientist launches several hypotheses in parallel, each agent running the same skill or different skills against different parameters. Each agent produces its own report, and results are aggregated for comparison. Autonomous does not mean unsupervised: the scientist reviews the plan and stays in control of key steps.

Campaigns are in development and currently available to a limited set of users. Contact your Owkin representative to find out more.

In none of these modes does the scientist need to write code or specify which datasets to query. K-Pro makes those decisions and shows them: the Thought process section of each answer lists the tools it called, and the answer states the data and methods used.

What's behind K-Pro

K-Pro draws on Owkin's patient data network: 166 partner institutions in more than 20 countries, representing over 24.5 million patients, with multimodal data covering the latest standard of care (bulk, single-cell, and spatial RNA, pathology, and longitudinal follow-up). Indications covered include oncology, immunology & inflammation, cardiometabolic, and neuro.

Agents & tools in K-Pro

Each question is routed to the agents and tools best suited to handle it. You don't need to choose them yourself.

Agent / tool	What it does
Plotter	Analyzes patient datasets and generates interactive plots and tables: cohort characterization, survival analysis, differential expression, pathway enrichment, and more, across clinical, bulk RNA-seq, WES, proteomics, single-cell, and spatial data.
BioKnowledge Navigator	Queries curated biological knowledge bases (such as Open Targets, ChEMBL, DepMap, the Human Protein Atlas, UniProt, and MSigDB) for gene- and target-level information: function, localization, essentiality, tractability, safety, and pathways.
Consensus	AI-powered academic search over more than 220 million peer-reviewed papers, for literature-backed answers with citations.
Web search	Searches the web for recent information, such as news or public trial and company information, with links to sources.
Pathology Explorer	Explores H&E slides (for example TCGA): whole-slide viewing, HistoPLUS cell-type maps, slide-level histomics, and survival analysis on cell-type features.
Reporter	Compiles the analyses from a conversation into a report you can export as a PDF.
Code execution & file upload	Runs custom analyses in a secure, isolated workspace, including on files you upload.
Trials Navigator	Explores the clinical trial landscape, with timelines, filters, and summary statistics.

What is a skill?

A skill is a pre-packaged set of instructions and tools designed for a high-stakes task. When you trigger a skill, K-Pro switches into a specific mode where it applies the right filters, checks the right thresholds, and returns a defined output format. Skills make analyses reproducible: the same workflow runs the same way every time, and skills are versioned so you can track changes.

Three types of skills are available:

• Owkin-certified skills: validated scientific workflows built from Owkin projects.
• KOL-derived skills: domain experts' methods and reasoning encoded as executable workflows.
• Your own skills: your internal best practices, layered on top of the existing library.

Skills are available per organization, on request. See Customize K-Pro to learn how to create, version, and share skills, and Writing reproducible skills for best practices.