End-to-end multimodal deep learning for real-time decoding of months-long neural activity from the same cells

AutoSort is designed to tackle two significant challenges in long-term stable recording.

• First, it efficiently aligns neurons over the course of long-term recordings to ensure consistent tracking of the same neurons each day.
• Second, it accurately sorts spikes while maintaining the precision throughout the recordings, ensuring that the performance achieved at the first of the recordings is sustained throughout the later days.

AutoSort innovatively leverages multimodal features as inputs. We extract single-channel waveform, multi-channel waveform, and the inferred spatial location for any potential spike that exceeds a certain threshold on any particular channel to be sorted.

For more details, please check out our publication.

• Read our tutorials with provided datasets .

Code used in the study is uploaded to figshare and will be public with the manuscript publication.

AutoSort package requires a standard computer with GPU to support the in-memory operations.

This package is supported for Linux. The package has been tested on the following system:

• Linux: Ubuntu 20.04

AutoSort mainly depends on the Python scientific stack.

spikeinterface
scipy
scikit-learn
pandas
pytorch
seaborn

pip install autosort-neuron==0.0.1.4

If you find AutoSort useful for your work, please cite our paper:

End-to-end multimodal deep learning for real-time decoding of months-long neural activity from the same cells. Yichun He#, Arnau Marin-Llobet#, Hao Sheng, Ren Liu, Jia Liu*. Preprint at bioRxiv (2024): https://doi.org/10.1101/2024.10.14.618046.