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

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatial transcriptomics methods are crucial for understanding tissue organization but often face scalability limitations when measuring individual genes.
• Emerging techniques like CISI, FISHnCHIPs, and ATLAS use aggregate transcriptional signatures to enhance throughput, but require careful feature design for effective integration with single-cell RNA sequencing (scRNA-seq).
• Optimizing only decoding accuracy in isolation neglects crucial experimental constraints, limiting the performance of aggregate measurement strategies.

Purpose of the Study:

• To develop a computational framework, CIPHER (Cell Identity Projection using Hybridization Encoding Rules), that jointly optimizes the design of aggregate transcriptional signatures and the downstream cell-type embedding.
• To integrate physical limitations of imaging assays directly into the optimization process to maximize discriminability and robustness to noise.
• To enable systematic, scRNA-seq-aligned feature design for scalable spatial transcriptomics using aggregate measurements.

Main Methods:

• CIPHER employs a neural-network framework to jointly optimize the experimental encoding matrix and the cell-type embedding.
• The framework incorporates physical constraints of imaging assays into its loss function, shaping the latent space for improved discriminability and robustness.
• A large-scale mouse brain scRNA-seq reference dataset was used to train and validate the model.

Main Results:

• CIPHER-designed encodings resulted in latent spaces with enhanced cell-type separability compared to existing methods.
• The framework demonstrated more uniform signal utilization and increased resilience to hybridization variability.
• Higher decoding accuracy was achieved in both simulated and experimental datasets, validating CIPHER's effectiveness.

Conclusions:

• CIPHER provides a principled approach for designing aggregate transcriptional signatures for scalable spatial transcriptomics.
• The joint optimization of decoding accuracy and experimental measurability addresses the feature design challenge in aggregate measurement strategies.
• CIPHER enables efficient and accurate reconstruction of cellular transcriptomes in spatial transcriptomics experiments.