POCO: Scalable Neural Forecasting through Population Conditioning.

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Machine Learning for Neuroscience

Background:

• Predicting future neural activity is crucial for understanding brain dynamics and developing neurotechnologies.
• Existing models often focus on interpretability or decoding, leaving neural forecasting, especially across sessions, underexplored.

Purpose of the Study:

• To introduce POCO, a unified model for accurate and generalizable neural forecasting across multiple recording sessions.
• To capture both neuron-specific and brain-wide dynamics for improved prediction of neural activity.

Main Methods:

• Developed POCO, a model combining a univariate forecaster with a population encoder.
• Trained POCO on five diverse calcium imaging datasets from zebrafish, mice, and C. elegans.
• Evaluated POCO's performance and adaptability with minimal fine-tuning on new recordings.

Main Results:

• POCO achieved state-of-the-art accuracy in predicting neural activity at cellular resolution during spontaneous behaviors.
• Learned embeddings from POCO recovered biologically meaningful structures, like brain region clustering, without anatomical labels.
• Identified key factors influencing performance, including context length, session diversity, and preprocessing.

Conclusions:

• POCO offers a scalable and adaptable approach for cross-session neural forecasting, applicable across species.
• The model provides actionable insights for designing future neural forecasting models.
• POCO lays the foundation for adaptive neurotechnologies and large-scale neural foundation models.