Peer-reviewed
-
Hosseini, M., Klein, I., Wunderle, V., Haustein, M., Semmler, C., Kramer, A., Tolve, M., Mardare, V., Galvao, A., Kuzu, T. D., Grefkes, C., Korotkova, T., Bueschges, A., Fink, G. R., Weiss, P. H., Daun, S., & Gatto, G. (2026). Cross-species identification of conserved and divergent locomotor kinematic strategies using AutoGaitA. Proceedings of the National Academy of Sciences, 123(35), e2534093123.
To compare how the nervous system coordinates movement across species, perturbations and behaviours, we developed AutoGaitA, an open-source Python toolbox for automated gait analysis. Analysing limb kinematics with AutoGaitA, we found that flies, mice and humans generate propulsive strength through different mechanisms but share a distal-to-proximal gradient of joint movement velocities. Ageing reduced propulsive strength in all three species while preserving this gradient, and in mice, locomotion adapted to ageing and task difficulty as an integrated function of both.
-
Hosseini, M., Zivony, A., Eimer, M., Wyble, B., & Bowman, H. (2024). Transient attention gates access consciousness: Coupling N2pc and P3 latencies using dynamic time warping. The Journal of Neuroscience, 44(26), e1798232024.
https://doi.org/10.1523/JNEUROSCI.1798-23.2024The N2pc and P3 event-related potentials index selective attention and access to working memory and conscious awareness, but a direct temporal link between them had not been convincingly shown. Using a new method based on Dynamic Time Warping (DTW), we analysed EEG from participants who reported target digits in rapid streams of letters and digits, and found that N2pc and P3 latencies were correlated, with the link being weaker when a distractor was reported. We provide this DTW-based method in the MATLAB and Python toolbox NeuroWarp, which can be used for time series of any kind.
-
Hosseini, M., Powell, M., Collins, J., Callahan-Flintoft, C., Jones, W., Bowman, H., & Wyble, B. (2020). I tried a bunch of things: The dangers of unexpected overfitting in classification of brain data. Neuroscience & Biobehavioral Reviews, 119, 456–467.
https://doi.org/10.1016/j.neubiorev.2020.09.036Machine learning helps neuroscientists extract information from EEG, fMRI and MEG data, but adjusting analysis settings after seeing the results can make findings partly or entirely spurious, a problem we call “overhyping”. We show that such spurious results can be obtained even from random data by changing hyperparameters in seemingly harmless ways, and that cross-validation does not protect against this. To prevent overhyping, we recommend safeguards already common in computer science and physics, such as lock boxes, blind analyses, pre-registration and nested cross-validation.
Under review
-
Hosseini, M., Zivony, A., Eimer, M., Wyble, B., & Bowman, H. (2026). The 2-feature Simultaneous Type/Serial Token Model of temporal feature conjunctions: Successes are merely fortunate conjunctions. bioRxiv (in review at Neural Networks).
https://doi.org/10.64898/2026.09.07.749788When people watch rapid streams of visual stimuli, they sometimes bind features of different items together and report a distractor instead of the target, an error known as a distractor intrusion. We present the 2f-ST2 model, the newest member of the Simultaneous Type/Serial Token family of neural models (e.g. Bowman & Wyble, 2007), which simulates the key behavioural and electrophysiological findings on distractor intrusions across diverse experimental paradigms without changing its architecture or core parameters. Because it relies on a single binding mechanism, the model is more parsimonious than the leading focused-attention/sophisticated-guessing account and implies that correct reports are simply fortunate conjunctions.
-
Jefferis, T., Nayak, A., Hosseini, M., Dogan, C., Ramaswamy, P., & Bowman, H. Amplitude and jitter biases in jackknife ERP latency estimates. In review at NeuroImage.
The jackknife method is widely used to compare the timing of event-related potentials (ERPs) between conditions, by measuring latencies on Grand Averages that each leave out one participant. We examine how these latency estimates are biased by differences in ERP amplitude and provide Dynamic Time Warping as a robust alternative.
Preprints
-
Tutas, J., Tolve, M., Özer-Yildiz, E., Ickert, L., Klein, I., Hosseini, M., Liebsch, F., Dethloff, F., Gavalisco, P., Endepols, H., Yang, M., Georgomanolis, T., Neumaier, B., Drzezga, A., Daun, S., Schwarz, G., Thorens, B., Gatto, G., Frezza, C., & Kononenko, N. (2024). Autophagy regulator ATG5 preserves cerebellar function by safeguarding its glycolytic activity. bioRxiv, 2024.01.27.577372.
https://doi.org/10.1101/2024.01.27.577372