> For the complete documentation index, see [llms.txt](https://docs.owkin.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.owkin.com/governance-and-security/model-governance/explainability-and-traceability/on-k-pro-skills.md). # On K Pro skills K Pro does not offer a single static way of discovering and prioritizing new target hypotheses. Instead, it provides a broad spectrum to enable both rule-based discovery/ranking of target lists and AI-driven ("self-drive") modes. Much effort of the Owkin team has been invested in deriving features from multimodal data, enabling the representation of targets in the respective feature space, and the relative ranking of targets in respective representations. On top of that, Owkin has developed different workflows (implemented as agentic skills) that utilize these features to generate ranked lists of novel target hypotheses. Most of these workflows aim to at the least cover typical rules used at pharma for target discovery, such as the ADC target discovery exemplified above. In addition to biological evidence, K Pro's workflows also consider competitive landscapes regarding each target hypothesis, population sizing (relevant for market sizing), druggability/tractability, assayability, and other, context dependent factors (e.g. tumor essentiality / addiction in the case of cell-intrinsic, i.e. oncogenic signaling-related targets). Notably, K Pro's workflows are editable/extendable/editable by the user, requiring no coding expertise due to the fact they are written in natural language. In addition to the rules-based approach, K Pro is currently being developed in the direction of a self-driving AI scientist: after light prompting by the user, it uses data and prior knowledge to generate hypotheses and iteratively engages more and orthogonal evidence sources to deepen, reject or prioritize those, generating structured reports for pharma scientists. In these autonomous discovery campaigns, K Pro taps on the entirety of knowledge, data, tools, and agentic skills to build the workflows in run-time. Importantly, the autonomous AI scientist capabilities are not meant to replace but to complement K Pro's rule-based discovery capabilities to offer the user a broader spectrum of capabilities and user control. **K Pro's scoring is hybrid and skill-based, not a single black-box ML model.** * **Skills are the unit of methodology.** A target prioritization skill (e.g. "ADC target prioritization for indication X") is a versioned package. It defines which features to compute, which data sources to query, which tools and models to call, and how to aggregate the results into a ranking. Skills are composable and inspectable. The methodology is auditable. * **Per-skill features** handle things like differential expression, healthy-tissue expression rank, malignant-cell specificity score, gene essentiality from DepMap, HistoPLUS-derived cell-type composition from H\&E, spatial co-localization scores from spatial transcriptomics etc. * **Aggregation uses weighted ranking and/or data-driven feature selection.** Recent internal work on ADC target prioritization used expert-based weighting given the scarcity of ground truth labels, but past work on small molecules implemented end-to-end data-driven weighting of the features and ranking of the targets. Weights are configurable per skill and per use case, and can be customized for customer strategic preferences (different weights for ADC vs. small-molecule vs. radioligand targets, for example). * **User customization sits on top of this.** Users can adjust weights at run time ("downweight literature, upweight spatial heterogeneity"). They can add custom constraints ("exclude any target with moderate or higher CNS expression"). They can register customer-internal scoring dimensions through custom MCP servers. They can customize existing skills or author new ones. As a consequence, we do not have one and only proprietary scoring algorithm that produces a calibrated success probability across all use cases. The scoring is the composition of explicit, inspectable skills. This is a strength for auditability and for working inside customer existing scientific frameworks. It is intentionally not a black box. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.owkin.com/governance-and-security/model-governance/explainability-and-traceability/on-k-pro-skills.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.