scLEMBAS: Context-Aware Modeling of Signaling Pathway Activity at Single-Cell Resolution

Area of Science:

• Computational Biology
• Systems Biology
• Molecular Biology

Background:

• Cells integrate extracellular cues via intracellular signaling networks, influencing transcription factor activity and cellular responses.
• Signaling pathways are complex, exhibiting non-linearity, feedback, and crosstalk, making them difficult to decipher.
• Context-dependent cellular responses to identical stimuli necessitate advanced computational modeling for prediction and interpretation.

Purpose of the Study:

• To present scLEMBAS, a context-aware, gray-box neural network designed to model signaling pathway activity at single-cell resolution.
• To bridge predictive capabilities of computational models with mechanistic interpretability using prior knowledge networks.
• To enable quantitative dissection of how signaling pathway activity is reshaped by perturbation within specific cellular contexts.

Main Methods:

• scLEMBAS encodes a prior-knowledge protein-protein interaction network as a recurrent neural network with learnable edge weights representing signaling interaction strengths.
• Context and cell variance are captured through compositional bias terms, and an adversarial approach facilitates counterfactual predictions.
• The model was validated on two scRNA-seq datasets covering single- and multi-perturbation scenarios.

Main Results:

• scLEMBAS accurately predicts out-of-distribution combinations of perturbation and context, outperforming existing models.
• The model captures population variance, enabling prediction of cell subtype-specific perturbation responses without explicit labels.
• Learned parameters, including edge weights and categorical biases, offer biological interpretability, identifying key proteins and interactions.

Conclusions:

• scLEMBAS provides a powerful framework for dissecting context-dependent cellular signaling and perturbation responses.
• The model enhances mechanistic understanding by preserving biological grounding within a predictive computational framework.
• scLEMBAS advances the ability to predict and interpret cellular behavior in complex biological systems.