Case studies library

Case study 1: Multimodal target & population characterization for asset positioning

The brief was to characterize two candidate targets for an IO asset in a solid-tumor indication: expression profile, intra-tumoral heterogeneity, spatial behavior in the tumor microenvironment, and outcome signal in the IO-treated subgroup.

The data substrate was Owkin's MOSAIC cohort for the indication, covering 490 real-world patients across five treatment cohorts, with 208 patients having all four modalities (bulk RNA-seq, single-cell RNA-seq, spatial transcriptomics, clinical follow-up).

The deliverable integrated all four modalities in a single report. Cross-modality concordance gave the client more confidence than any single modality could provide. Intra-tumoral heterogeneity and spatial autocorrelation surfaced patterns that bulk-only platforms cannot produce.

The skills underlying this analysis (cohort building, multimodal target characterization, spatial TME analysis, and outcome association) are the same skills K Pro orchestrates as composable workflows today.

---

Case study 2: Drug to clinically actionable patient subgroups

The brief was to identify the patient subgroup most likely to respond to an immune checkpoint asset (an inhibitory NK-receptor antagonist combined with an anti-PD1) across two solid-tumor indications, and to distill the answer into clinically implementable selection rules a clinician could apply at trial screening.

Approach. A drug-relevant feature set spanning the target axis, checkpoints, NK and T cell signatures, and the broader microenvironment was scored per patient. Consensus clustering yielded three subgroups along a gradient of drug-enabling context (cold, warm, hot). Cluster labels transferred to two independent private cohorts at Jaccard 0.87 and Cohen Kappa 0.87 (n=278 on the larger cohort). The subgroups were distilled into a three-gene transcriptomic signature and a histomics decision tree (PDL1 status, lymphocyte density, cancer-cell density, TLS) reaching sensitivity 0.80 and specificity 0.74 on H&E. Trial simulation estimated approximately 30% reduction in recruitment needed for an equivalent-powered Phase III in the larger indication. The client is now validating these biomarkers prospectively in Phase II.

K Pro's drug-positioning skill suite covers this exact five-step workflow: cohort building, drug-context feature engineering, patient clustering, cluster characterization, and biomarker distillation across transcriptomics, histomics, and clinical modalities.