PertSpectra: Interpretable Matrix Factorization for Predicting Functional Impact of Genetic Perturbation Experiments.

Area of Science:

• Genomics
• Systems Biology
• Drug Discovery

Background:

• Genetic perturbations are crucial for understanding disease mechanisms.
• Interpreting combinatorial perturbation data remains a significant challenge in drug discovery.
• Current methods struggle with the complexity of large-scale genetic screens.

Purpose of the Study:

• To introduce PertSpectra, a novel computational method for analyzing genetic perturbation data.
• To improve the biological interpretation of complex combinatorial perturbation screens.
• To enhance the understanding of disease mechanisms through genetic perturbation analysis.

Main Methods:

• Developed PertSpectra, a guided triple matrix factorization technique.
• Incorporated perturbation information and gene-gene interaction graph priors.
• Applied the method to single-cell RNA sequencing datasets with various genetic perturbations.

Main Results:

• PertSpectra generates sparse, biologically relevant latent factors.
• The method demonstrates strong interpretability of the latent space.
• PertSpectra accurately predicts unseen perturbation combinations and stratifies similar perturbations.

Conclusions:

• PertSpectra offers an integrated approach for modeling combinatorial perturbation data.
• This method facilitates a deeper understanding of genetic effects in drug discovery.
• PertSpectra addresses key limitations in current perturbation analysis methodologies.