Opening the
Black Box.
In medicine, a prediction without a reason is dangerous. Ambr uses Explainable AI (XAI) to ensure that every risk score comes with a transparent, biological rationale.
Why "Black Box" AI Fails in Clinic
Traditional AI models process data through hidden layers indecipherable to humans. They output a probability (e.g., "85% risk") but cannot explain why.
Clinical Risk
"If a doctor cannot understand the basis of a recommendation, they cannot safely act on it. Trust is not about accuracy alone; it is about interpretability."
How the Algorithm is Built
Our framework moves from raw biological data to actionable insight through a rigorous, multi-stage pipeline.
Variable selection
Thorough litterature review from scientific and clinical research indentify the risk factors contributing to the prediction and diagnostic of a medical condition.
Model training
We built Machine Learning models using the selected variable, and trained them on more than 550.000 individuals to identify the interactions and weight of those factors.
Validation
The models were trained on global datasets. We then used various cross-validation to prevent overfitting and ensure the model generalizes to new patients.
Explainability
Instead of stopping at the prediction, we apply SHAP (SHapley Additive exPlanations) to reverse-engineer exactly how much each factor contributed to the result.
SHAP: Biological Attribution
This game-theoretic approach assigns a "contribution score" to every single feature in your patient's profile.
Instead of a single risk number, we show you the specific biomarkers driving that risk up or down. This transforms the algorithm from an oracle into a transparent informations partner.
Risk Factor Decomposition
Patient ID: #AMB-8921 // Predicted Risk: High (82%)
The high Lp(a) is the primary driver. Despite good VO2 Max (protective), the genetic lipid risk outweighs lifestyle benefits.