AKLIMATE

AKLIMATE integrates multi-omics data with pathway and gene-signature priors using a stacked multiple-kernel learning framework to enable predictive modeling and interpretation of phenotypes, survival outcomes, and gene knockdown responses.

Key Features:

• Multi-Omics Integration: Incorporates diverse omic datasets such as genomics and transcriptomics together with pathway information for comprehensive analyses.
• Prior-Knowledge Feature Sets: Uses pathways and gene signatures as prior knowledge to define biologically coherent feature sets.
• Stacked Multiple-Kernel Learning: Employs a stacked multiple-kernel learning framework to combine information from multiple kernels into a predictive model.
• Random-Forest–Derived Kernels: Constructs individual kernels from random forests trained on genes within specific pathways to capture prediction propensities.
• Cross-Platform Feature Linking via Pathways: Connects feature data across different platforms by mapping features to common pathways.
• Interpretability: Quantifies the importance of individual genes and pathways in model accuracy to provide interpretable results.

Scientific Applications:

• Phenotype Prediction: Predicts phenotypes including microsatellite instability in cancers such as endometrial and colorectal cancer.
• Survival Analysis: Supports survival prediction tasks, exemplified by applications in breast cancer.
• Gene Knockdown Response Prediction: Predicts cell line responses to gene knockdowns to aid identification of synthetic lethality and genetic interactions.

Methodology:

Integrates multi-omics data with pathway information via a stacked multiple-kernel learning algorithm in which each kernel is derived from random forests trained on genes within specific pathways to capture prediction propensities and link features across platforms.