PPubMed4 Mar 2024Cited 21×

A machine learning toolbox for the analysis of sharp-wave ripples reveals common waveform features across species

Andrea Navas-Olive, Adrian Rubio, Saman Abbaspoor, Kari L. Hoffman, Liset M. de la Prida

Abstract

The study of sharp-wave ripples has advanced our understanding of memory function, and their alteration in neurological conditions such as epilepsy is considered a biomarker of dysfunction. Sharp-wave ripples exhibit diverse waveforms and properties that cannot be fully characterized by spectral methods alone. Here, we describe a toolbox of machine-learning models for automatic detection and analysis of these events. The machine-learning architectures, which resulted from a crowdsourced hackathon, are able to capture a wealth of ripple features recorded in the dorsal hippocampus of mice across awake and sleep conditions. When applied to data from the macaque hippocampus, these models are able to generalize detection and reveal shared properties across species. We hereby provide a user-friendly open-source toolbox for model use and extension, which can help to accelerate and standardize analysis of sharp-wave ripples, lowering the threshold for its adoption in biomedical applications.

A figure from A machine learning toolbox for the analysis of sharp-wave ripples reveals common waveform features across species — fig. from the paper

§ The Valyu brief

The paper, in five minutes

Read from the full text via Valyu Contents

A new open-source ML toolbox accurately detects hippocampal sharp‑wave ripples in mice and macaques, outperforming traditional methods.

What’s new

i.Created an open‑source toolbox of five top‑performing machine‑learning models for sharp‑wave ripple detection.
ii.Organized a crowdsourced hackathon to select and benchmark ML architectures (XGBoost, SVM, LSTM, 2D‑CNN, 1D‑CNN).
iii.Demonstrated that the models generalize from mouse to macaque hippocampal recordings.
iv.Provided an ensemble model that combines individual architectures for improved performance.

How they did it

A hackathon generated 18 candidate models, from which five architectures were retrained on a z‑scored, down‑sampled (1250 Hz) dataset of 7586 events across 21 sessions (8 mice). Models were evaluated on a held‑out test set using precision, recall and F1‑score, and the best model per architecture was selected for comparison and ensemble construction.

What they found

50 best models were tested; the model with the highest mean F1 was chosen for between‑model comparison.
Kruskal‑Wallis tests showed significant effects of input channel number on F1 for XGBoost, SVM, LSTM, and 1D‑CNN (p < 0.001).
Ensemble model weights: w1 = ‑0.11 (XGBoost), w2 = ‑1.56 (SVM), w3 = 5.33 (LSTM), w4 = 2.03 (2D‑CNN), w5 = 4.07 (1D‑CNN), bias = ‑4.97.

§ The Valyu brief

Reading the full paper and taking notes. This takes a few seconds…

§ Ask this paper

Ask a question about this paper

Valyu reads the full text and answers from what the paper actually says.

§ Related work in other archives

A machine learning toolbox for the analysis of sharp-wave ripples reveals common waveform features across species

The paper, in five minutes

Ask a question about this paper

Similar papers in all four archives

Similar papers in all four archives

The paper, in five minutes