SpikeForest

SpikeForest benchmarks automated neural spike sorting algorithms against curated electrophysiological recordings with ground-truth units to evaluate sorter accuracy.

Key Features:

• Curated database: Contains 650 recordings (~1.3 TB) comprising approximately 35,000 ground-truth units.
• Diverse recording types: Includes extracellular recordings paired with intracellular voltages, simulated recordings, and hybrid synthetic datasets.
• Multi-laboratory contributions: Aggregates recordings contributed by over a dozen laboratories.
• Sorter integration: Evaluates ten modern spike sorting codes integrated under a unified Python framework.
• Automated benchmarking pipeline: Executes systematic evaluations on a compute cluster via an automated pipeline to enable reproducible benchmarking.
• Ground-truth validation: Performs ground-truth validation of automated spike sorters using the curated dataset.

Scientific Applications:

• Algorithm benchmarking: Benchmark performance of automated spike sorting algorithms across diverse datasets with ground truth.
• Algorithm validation: Validate spike sorting accuracy against intracellular ground-truth and synthetic ground-truth units.
• Method selection for experiments: Inform selection of spike sorters and sorter parameters for specific probes and brain regions.
• Comparative studies: Compare algorithm performance across experimental conditions and recording types.

Methodology:

Integration of multiple spike sorting codes under a unified Python framework and systematic evaluation of ten spike sorters on a compute cluster via an automated pipeline using the curated electrophysiological dataset.