Fine tuning a logical model of cancer cells to predict drug synergies: combining manual curation and automated parameterization

Area of Science:

• Systems biology
• Computational oncology
• Drug discovery

Background:

• Drug combinations offer promise for personalized disease therapy.
• Logical models, manually defined, can identify synergistic cancer signaling inhibitors.
• Manual model optimization is labor-intensive.

Purpose of the Study:

• To demonstrate automated adjustments of model topology and logic equations for logical model optimization.
• To explore larger model ensembles with less restraint where parameterization is not biologically guided.
• To benchmark synergy prediction performance of logical models in targeted drug combinations.

Main Methods:

• Developed an automated methodology for logical model optimization, adjusting topology and logic equations.
• Utilized a genetic algorithm for automated parameterization.
• Benchmarked model performance on a dataset of 153 targeted drug combinations and validated predictions experimentally.

Main Results:

• Automated adjustments significantly reduced workload compared to manual model definition.
• Automated parameterization using a genetic algorithm outperformed manual models in synergy prediction.
• Data-guided model refinement improved prediction quality, showing tolerance to some data errors but sensitivity to overall data quality.
• Predicted the synergistic effect of joint PI3K and TAK1 inhibition, validated in cancer cell cultures and xenograft experiments.

Conclusions:

• Automated logical modeling is a powerful tool for optimizing synergy prediction in targeted drug combinations.
• This framework facilitates the exploration of larger, more flexible model ensembles.
• Predictive logical models can propose and validate novel therapeutic mechanisms, such as combined PI3K and TAK1 inhibition.