Highly Accurate Cancer Phenotype Prediction with AKLIMATE, a Stacked Kernel Learner Integrating Multimodal Genomic Data and Pathway Knowledge

Abstract

Advancements in sequencing have led to the proliferation of multi-omic profiles of human cells under different conditions and perturbations. In addition, several databases have amassed information about pathways and gene "signatures" -- patterns of gene expression associated with specific cellular and phenotypic contexts. An important current challenge in systems biology is to leverage such knowledge about gene coordination to maximize the predictive power and generalization of models applied to high-throughput datasets. However, few such integrative approaches exist that also provide interpretable results quantifying the importance of individual genes and pathways to model accuracy. We introduce AKLIMATE, a first kernel-based stacked learner that seamlessly incorporates multi-omics feature data with prior information in the form of pathways for either regression or classification tasks. AKLIMATE uses a novel multiple-kernel learning framework where individual kernels capture the prediction propensities recorded in random forests, each built from a specific pathway gene set that integrates all omics data for its member genes. AKLIMATE outperforms state-of-the-art methods on diverse phenotype learning tasks, including predicting microsatellite instability in endometrial and colorectal cancer, survival in breast cancer, and cell line response to gene knockdowns. We show how AKLIMATE is able to connect feature data across data platforms through their common pathways to identify examples of several known and novel contributors of cancer and synthetic lethality.

Prior Biological Knowledge Improves Genomic Prediction of Growth-Related Traits in Arabidopsis thaliana

“…The approach using predicted functional information proved to be more useful in this context, but more approaches and sources of information can also be incorporated with a focus on prioritizing biologically related genomic regions. Moreover, knowledge from multiple heterogeneous sources can be combined to further pinpoint potential QTLs, termed as poly-omics GP models (Wheeler et al, 2014 ; Uzunangelov et al, 2020 ). These information sources may include (i) predicted variants effects, (ii) gene functions e.g., GO, COEX, (iii) networks of gene-gene and protein-protein interactions, stored in public resources like STRING (Mering et al, 2003 ), GeneMANIA (Warde-Farley et al, 2010 ); (iv) pathways, in which genes are grouped e.g., KEGG (Kanehisa and Goto, 2000 ); (v) previously generated GWAS and QTL results which indicate involvement of particular regions for specific traits e.g., AraGWAS (Togninalli et al, 2020 ), AraQTL (Nijveen et al, 2017 ), (vi) known connections to phenotypes and (vii) endophenotypes, usually measured using -omics data at different stages of genetic information flow toward phenotypes.…”

Knowledge-driven approaches to improve genomic prediction in plants

“…The approach using predicted functional information proved to be more useful in this context, but more approaches and sources of information can also be incorporated with a focus on prioritizing biologically related genomic regions. Moreover, knowledge from multiple heterogeneous sources can be combined to further pinpoint potential QTLs, termed as poly-omics GP models (Wheeler, Aquino-Michaels et al 2014, Uzunangelov, Wong et al 2020). These information sources may include (i) predicted variants effects, (ii) gene functions e.g.…”