Target identification & indication discovery

Use K Pro to identify and prioritize therapeutic targets, and to characterize the patient subgroups in which they're most relevant.

• Prioritize targets: rank therapeutic targets and gene candidates using statistical evidence from TCGA and MOSAIC datasets.

• Characterize patient subgroups: identify distinct populations based on multi-omics profiles and clinical outcomes.

The pages below walk through each task with a tested example prompt.

Here is a concrete worked example of how K Pro navigates target prioritization for Antibody-Drug Conjugates (ADCs) or, analogously, targeted alpha therapies (TATs). A similar workflow is used to identify radio-ligand therapies (RLTs), except that protein localization on the cell membrane is not required.

1. User request. A researcher prompts K Pro: "Identify and prioritize top ADC targets for bladder cancer (BLCA)."

2. K Pro reasons about the constraints. K Pro translates the query into a list of biological constraints and respective analyses. For a viable ADC target, the system knows the protein must be expressed on the cell surface, be highly prevalent in BLCA tumors, and have minimal (optimally, zero) expression in essential healthy tissues to ensure a safe therapeutic window. It also reasons that expression heterogeneity might drive resistance, and hence good targets are expressed in a high proportion of cells/regions of each tumor, and that non-expressing regions should optimally be impacted through the bystander effect.

3. K Pro analyses bulk + single-cell + spatial transcriptomics. The system analyzes bulk transcriptomics (TCGA, MOSAIC) as well as single-cell and spatial transcriptomics (MOSAIC) data, scoring each target according to the above criteria. Targets score highly if they are expressed highly in tumor and not expressed in normal tissue (using GTEx and single-cell normal tissue atlases), and expressed either ubiquitously within the tumor (all single cells / spatial regions) or expressed in a pattern that allows the bystander effect to kill nearby tumor cells that do not express the target (based on a bystander score feature Owkin has developed).

4. K Pro outputs targets ranked by quality. The system provides a series of both textual and visual (plots) outputs that provide target ranking/recommendations and per-target evidence for target quality.

5. Optional competitive intelligence. If opted & prompted by the user, the system can provide recommendations rooted not only in biological data but also on competitive landscape analysis. The system conducts in-depth competitive intelligence analysis (including trial data, patent data, news feeds, conference abstracts, etc.) through its competitive intelligence agentic capabilities.

Since this example is about expression-based targets (ADCs, TATs), some features typically associated with small molecules like genetic association, dependency/addiction, biochemical pathways, for example, are not relevant and hence the system correctly does not apply them.

At the end, the user can ask to download the analyses as a comprehensive report. Alternatively, the sequence of investigation can be bundled in a skill that helps reproduce the exact same flow without repeating all the sequential prompts.