Modeling and predicting single-cell multi-gene perturbation responses with scLAMBDA.

Area of Science:

• Computational Biology
• Genomics
• Molecular Biology

Background:

• Understanding cellular responses to genetic changes is key for gene regulation and phenotype research.
• Single-cell RNA sequencing (scRNA-seq) offers detailed insights but requires advanced computational tools for mechanism decoding and outcome prediction.
• Predicting perturbation effects is crucial for efficient experimental design.

Purpose of the Study:

• To introduce scLAMBDA, a deep generative learning framework for modeling and predicting single-cell transcriptional responses to genetic perturbations.
• To leverage large language models for integrating prior biological knowledge and disentangling cell states from perturbation effects.
• To provide a versatile tool for predicting outcomes of single-gene and combinatorial genetic perturbations.

Main Methods:

• Developed scLAMBDA, a deep generative learning framework utilizing gene embeddings from large language models.
• Integrated prior biological knowledge into the model to distinguish basal cell states from perturbation-specific representations.
• Evaluated scLAMBDA on multiple single-cell CRISPR Perturb-seq datasets.

Main Results:

• scLAMBDA consistently outperformed existing methods in predicting perturbation outcomes with higher accuracy.
• Demonstrated robust generalization to novel genes and perturbations.
• Successfully captured both average expression changes and the heterogeneity of single-cell responses.

Conclusions:

• scLAMBDA is a powerful computational tool for predicting cellular responses to genetic perturbations.
• Its predictions facilitate downstream analyses like identifying differentially expressed genes and exploring genetic interactions.
• The framework enhances experimental design by prioritizing target genes for genetic perturbation studies.