Explainable predictive and prognostic biomarkers: interpretable-by-design models for chronic diseases

Datharsis, a University of Granada spin-off specialised in explainable data analysis, is looking to join a consortium as the partner responsible for data analysis and integration, either leading the data analysis work package or contributing specific analysis tasks.

AI can predict disease progression and treatment response, but a prediction a clinician cannot question and a regulator cannot audit will not reach the clinic. Our models are interpretable by design, not explained afterwards with post-hoc layers. They show which variables drive each prediction and how confident it is, which is what clinical validation, regulatory submission and clinical adoption require.

What we offer:

• Statistical design of the study from the proposal stage: factors, cohort structure and sample size, to reach the required statistical power with fewer patients

• Integration of omics layers (metabolomics, proteomics, transcriptomics, epigenomics, metagenomics, spatial omics) with clinical, imaging-derived and real-world data

• Discovery of interpretable biomarkers and signatures that work on small or unbalanced cohorts, where most AI approaches overfit

• Rigorous validation with cross-validation and permutation testing, to separate real signal from artefacts

• Reproducible, auditable, FAIR-aligned analysis pipelines

• Access to DathaSuite, our analysis platform, and training for consortium researchers through our DathaXplain programme

Our team’s methods have been applied to metabolomic prediction of chemotherapy response in breast cancer, metabolomic survival signatures in colorectal liver metastasis, and multi-omics profiling of a paediatric cohort combining microbiota, metabolomics and clinical data. See these and other documented cases.

We are looking for coordinators and partners with clinical cohorts, registries or patient data who need a dedicated partner for interpretable, validated analysis.