The missing link: Predicting connectomes from noisy and partially observed tract tracing data.

Area of Science:

• Neuroscience
• Computational Biology
• Systems Neuroscience

Background:

• Understanding the brain's wiring map, or connectome, has advanced due to neuroimaging and computational tools.
• Establishing a definitive connectome for any species remains challenging, with tracer studies being reliable but costly and ex vivo.
• Existing data limitations necessitate predictive approaches for mapping neural connections.

Purpose of the Study:

• To develop and validate a predictive model for identifying unknown neural connections within a connectome.
• To leverage existing connectivity data to infer unobserved neural pathways.
• To assess the model's performance against established methods and explore its utility in integrating multimodal data.

Main Methods:

• A 'latent space model' was employed to embed neural connectivity into an abstract physical space.
• The model learns embeddings from observed connections to predict unobserved ones.
• The methodology was applied to macaque connectivity datasets and integrated multimodal data for the mouse neocortex.

Main Results:

• The latent space model successfully predicted unobserved neural connectivity in macaque datasets.
• The model outperformed two baseline methods and an alternative model in predictive accuracy.
• The approach demonstrated the ability to integrate multimodal observations (anterograde and retrograde tracers).

Conclusions:

• The latent space model offers a powerful, data-driven approach to predict neural connections and complete the connectome.
• This method overcomes limitations of traditional tracer studies by predicting, rather than solely probing, connections.
• The probabilistic nature of the model highlights predictable and difficult connections, guiding future experimental investigations.