New multimodal AI system predicts drug combination safety from lab data before clinical trials

Researchers have developed an AI system capable of forecasting how drug combinations will perform in patients by analyzing preclinical laboratory and molecular data. The approach could help pharmaceutical companies avoid costly late-stage trial failures by identifying risky drug pairings earlier in development.

How Madrigal works#

The new model, called Madrigal, integrates multiple types of information into a unified framework. Rather than relying solely on a drug's chemical structure and known targets—the typical approach—Madrigal incorporates how compounds actually behave in cells: their effects on cell viability and changes to gene expression patterns (transcriptomics). The system analyzes data from across 21,842 compounds and learns to map all these different data types into a shared mathematical space, allowing it to make predictions even when a drug has been tested in only some of the available assays.

Validation against clinical evidence#

The researchers trained Madrigal on 158 expert-reviewed drug combinations with documented clinical outcomes and 795 patient-reported adverse events from real treatments. When tested on head-to-head clinical trials comparing two treatment arms, the model correctly ranked the safer arm in 25 of 28 comparisons based on four serious side effects: neutropenia, anemia, hair loss, and low blood sugar.

The system also showed predictive power in a separate oncology patient cohort and identified treatment effectiveness in acute myeloid leukemia samples. In metabolic disease research, Madrigal ranked resmetirom favorably for safety when combined with type 2 diabetes medications.

Why this matters#

Drug combinations are increasingly central to modern treatment, but predicting their safety in advance remains difficult. Synergistic toxicity—where two drugs interact in unexpected ways—has long been a source of clinical failures. By catching these interactions earlier, using only data from laboratory experiments rather than waiting for human trials, developers can redirect resources toward more promising candidates and get safer medicines to patients faster.