Methodology

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.

The Conflict

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."

Input Layer
Genetics
Proteomics
Lifestyle
History
?
Hidden Processing
Output
High Risk
Reason: Unknown
Create evidence (hallucination)
Architecture

How the Algorithm is Built

Our framework moves from raw biological data to actionable insight through a rigorous, multi-stage pipeline.

Step 01

Variable selection

Thorough litterature review from scientific and clinical research indentify the risk factors contributing to the prediction and diagnostic of a medical condition.

Step 02

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.

Step 03

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.

Step 04

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.

The Solution

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%)

Analysis Type
SHAP Force Plot
Lp(a) Levels
+35%
Sleep Duration
+12%
VO2 Max
-20%
Homocysteine
+2%
Interpretation:

The high Lp(a) is the primary driver. Despite good VO2 Max (protective), the genetic lipid risk outweighs lifestyle benefits.