Related Experiment Video
Updated: Dec 27, 2025
09:33
Author Spotlight: Finding New Therapeutic Targets for Malignant Peripheral Nerve Sheath Tumor Through Genome-Scale shRNA Screens
Published on: August 25, 2023
1.5K
KCML: a machine-learning framework for inference of multi-scale gene functions from genetic perturbation screens
Heba Z Sailem1,2, Jens Rittscher1,2, Lucas Pelkmans3
1Department of Engineering Science, Institute of Biomedical Engineering, University of Oxford, Oxford, UK.
Molecular Systems Biology
|March 7, 2020
Summary
We developed Knowledge- and Context-driven Machine Learning (KCML) to predict gene functions in different contexts. KCML improves upon traditional methods by systematically identifying context-specific gene roles in health and disease.
Area of Science:
- Genomics
- Systems Biology
- Bioinformatics
Background:
- Understanding context-dependent gene functions is vital for deciphering the genetic underpinnings of health and disease.
- Current methods for inferring gene functions from large-scale genetic perturbation screens rely on ad hoc analysis pipelines.
- These traditional pipelines often involve unsupervised clustering and functional enrichment, which may lack systematic context-specificity.
Purpose of the Study:
- To introduce Knowledge- and Context-driven Machine Learning (KCML), a novel framework for systematically predicting multiple context-specific gene functions.
- To demonstrate KCML's ability to predict gene functions based on the similarity of perturbation phenotypes to those with known functions.
- To showcase KCML's generalizability across different biological scales (molecular, cellular, population) and its applicability to various genetic perturbation screens.
Main Methods:
- Developed the Knowledge- and Context-driven Machine Learning (KCML) framework.
- Applied KCML to three distinct datasets representing molecular, cellular, and population-level phenotypes.
- Compared KCML's performance against traditional ad hoc analysis pipelines for gene function inference.
Main Results:
- KCML significantly outperforms traditional analysis pipelines in predicting context-specific gene functions.
- KCML identified an abnormal multicellular organization phenotype linked to olfactory receptor depletion and TGFβ/WNT signaling genes in colorectal cancer cells.
- Validation in colorectal cancer patients confirmed that olfactory receptor expression predicts poorer patient outcomes.
Conclusions:
- KCML provides a systematic framework for discovering novel, scale-crossing, and context-dependent gene functions.
- The framework is highly generalizable and applicable to a wide range of large-scale genetic perturbation screens.
- The findings underscore the importance of context in gene function and disease, with potential implications for cancer research and patient stratification.
