CIPHER: An end-to-end framework for designing optimized aggregated spatial transcriptomics experiments.

Area of Science:

• Spatial transcriptomics
• Computational biology
• Genomics

Background:

• Current imaging-based spatial transcriptomics methods often measure individual genes, limiting scalability and requiring integration with single-cell RNA sequencing (scRNA-seq) for comprehensive cellular state analysis.
• Emerging techniques like CISI, FISHnCHIPs, and ATLAS use aggregate transcriptional signatures to enhance throughput, but this shifts the challenge to designing effective features.
• Optimal aggregate signatures must be discriminative, measurable, robust to noise, and possess balanced signal and dynamic range for successful scRNA-seq integration.

Purpose of the Study:

• To develop a novel framework, CIPHER (Cell Identity Projection using Hybridization Encoding Rules), for optimizing the design of aggregate transcriptional signatures in spatial transcriptomics.
• To jointly optimize the experimental encoding matrix (gene aggregation) and the downstream cell embedding within a unified neural network framework.
• To enhance the accuracy and robustness of cell type identification in spatial transcriptomics by accounting for experimental limitations.

Main Methods:

• CIPHER employs a neural network framework that jointly optimizes the gene-to-signature aggregation (encoding matrix) and the cell embedding process.
• The framework directly incorporates physical imaging assay limitations into its loss function, guiding the latent space optimization.
• This approach aims to maximize cell-type discriminability while ensuring robustness against measurement noise and signal constraints.

Main Results:

• CIPHER-designed encodings demonstrate improved cell-type separability in latent space representations.
• The framework promotes uniform signal utilization across signatures and enhances resilience to hybridization variability.
• CIPHER achieves higher decoding accuracy for cell types using both simulated and experimental spatial transcriptomics data, validated against a mouse brain scRNA-seq reference.

Conclusions:

• CIPHER presents a novel approach to aggregate signature design by framing it as a joint optimization problem balancing decoding accuracy and experimental measurability.
• This method enables systematic and scRNA-seq-aligned feature design for scalable spatial transcriptomics utilizing aggregate measurements.
• CIPHER facilitates more accurate and robust cell type identification in spatial transcriptomics, overcoming limitations of previous methods.