Protocol for using Ciclops to build models trained on cross-platform transcriptome data for clinical outcome...

Area of Science:

• Computational biology
• Biostatistics
• Machine learning in healthcare

Background:

• Predicting clinical outcomes using cross-platform data presents challenges in model robustness and generalizability.
• Explainable models are crucial for understanding performance variations due to sample conditions.

Purpose of the Study:

• To present Ciclops, a freely available software for developing explainable predictive models.
• To enable robust clinical outcome predictions across diverse, cross-platform datasets.
• To integrate SHAP analysis for biomarker discovery.

Main Methods:

• Utilized Ciclops software for cross-platform training of predictive models.
• Employed SHAP (SHapley Additive exPlanations) for post-training analysis.
• Validated model performance and explainability on cross-platform clinical datasets.

Main Results:

• Ciclops successfully builds explainable models for clinical outcome prediction.
• The developed models demonstrate generalizability across different datasets.
• SHAP analysis effectively identified potential biomarkers associated with clinical outcomes.

Conclusions:

• Ciclops provides a valuable tool for creating reliable and interpretable predictive models in clinical research.
• The approach facilitates biomarker discovery and enhances the understanding of model behavior.
• This method improves the application of machine learning to cross-platform clinical data.