Periodic Reporting for period 1 - SubcorticalBCI (A subcortical BCI to restore skilled movement)
Reporting period: 2021-06-01 to 2023-05-31
Brain-computer interfaces (BCIs) read the user’s intent directly from the brain activity. Hence, they provide a unique opportunity to restore movement to patients with paralysis or reduced ability to move because of spinal cord injury, stroke, or neuro-degenerative disease such as Parkinson’s. Indeed, in animal models and in humans, BCIs have restored some voluntary movement through bypassing neurological injuries. However, BCIs still face major challenges, most notably, providing accurate interface that enables seamless and naturalistic control for different users without the need for extensive training. Here, I adopted a multidisciplinary approach, integrating recent advances in computational tools in neuroscience and large-scale neural recording datasets to demonstrate that brain activity is similar across different animals as long as the behaviour of the animals is similar. This similarity in brain activity allowed us to predict the behaviour of one animal using a decoder built using data from a different animal, thereby avoiding the need to recalibrate our model for every individual. Our achievement in this project has the potential to significantly enhance the usability of BCIs by making them ‘work out of the box’ without the need to tailor them to each individual user.
The work I performed could be divided into two parts. My efforts toward performing experiments in my host lab, and my work on validating the computational methods required for the project that I disseminated in multiple conferences and published in the prestigious scientific journal, Nature. Following I will review both aspects in details.
Due to the shortages and delays caused by the global pandemic, I had to spend much longer than expected on equipping my host lab and building the setup. The experiments were further delayed because acquiring authorisations for animal work in the UK were backlogged. We managed to start the experimental work only in mid-2022. I have since improved the experimental rig and recently started training animals in the task proposed in this project. Piloting various aspects of the behavioural task and electrophysiological recordings is ongoing and in the next few months, I will train the first batch of animals in the task and address the central hypothesis of this project.
To mitigate the impact of the pandemic and other issues outlined above, I decided to develop the computational tools required for this project while the experiments were postponed. Through collaborations, we acquired datasets in monkeys and mice performing different motor tasks. I analysed the electrophysiological and behavioural data to study the similarity of neural dynamics across different animals performing similar tasks. This technique will be directly applicable to the future BCI task. Generalising a BCI across subjects is a main goal of this project —and a landmark contribution to the field— and will accelerate the BCI experiments I am currently piloting.
We demonstrated that animals of the same species exhibit remarkably similar neural population dynamics when performing similar behaviours, despite the individual variations in their neural circuitry. Using neural recordings from monkeys and mice engaged in upper limb tasks, we employed Principal Component Analysis (PCA) and Canonical Correlation Analysis (CCA) to uncover and subsequently align the coordinated activity of large populations of single neurons (the so-called latent dynamics). We found that these dynamics are not only preserved across individuals but are also behaviourally relevant, allowing for the decoding of intended and ongoing movements. Furthermore, these preserved dynamics extend beyond the cortex to the dorsal striatum and are present even during the planning phase of movements. We also found that the extent of preservation in neural dynamics was directly related to the stereotypy of the behaviour. Finally, we could train recurrent neural networks that produced the same behaviour with markedly different dynamics. This showed that behavioural similarity is not sufficient for observing similar neural dynamics.
I participated in three conferences to present and discuss my results at different stages of the project. Namely, I presented in FENS Forum 2022 (organised by the Federation of European Neuroscience Societies), NCM 2022 (organised by the Society for the Neural Control of Movement), and in iBAGS 2023 (organised by the International Basal Ganglia Society). This work was finally published as a pre-print in September 2022, and its final revised version is also published open access in Nature. The entire codebase to reproduce every figure and most of the datasets used in the paper are also available online.
Due to the shortages and delays caused by the global pandemic, I had to spend much longer than expected on equipping my host lab and building the setup. The experiments were further delayed because acquiring authorisations for animal work in the UK were backlogged. We managed to start the experimental work only in mid-2022. I have since improved the experimental rig and recently started training animals in the task proposed in this project. Piloting various aspects of the behavioural task and electrophysiological recordings is ongoing and in the next few months, I will train the first batch of animals in the task and address the central hypothesis of this project.
To mitigate the impact of the pandemic and other issues outlined above, I decided to develop the computational tools required for this project while the experiments were postponed. Through collaborations, we acquired datasets in monkeys and mice performing different motor tasks. I analysed the electrophysiological and behavioural data to study the similarity of neural dynamics across different animals performing similar tasks. This technique will be directly applicable to the future BCI task. Generalising a BCI across subjects is a main goal of this project —and a landmark contribution to the field— and will accelerate the BCI experiments I am currently piloting.
We demonstrated that animals of the same species exhibit remarkably similar neural population dynamics when performing similar behaviours, despite the individual variations in their neural circuitry. Using neural recordings from monkeys and mice engaged in upper limb tasks, we employed Principal Component Analysis (PCA) and Canonical Correlation Analysis (CCA) to uncover and subsequently align the coordinated activity of large populations of single neurons (the so-called latent dynamics). We found that these dynamics are not only preserved across individuals but are also behaviourally relevant, allowing for the decoding of intended and ongoing movements. Furthermore, these preserved dynamics extend beyond the cortex to the dorsal striatum and are present even during the planning phase of movements. We also found that the extent of preservation in neural dynamics was directly related to the stereotypy of the behaviour. Finally, we could train recurrent neural networks that produced the same behaviour with markedly different dynamics. This showed that behavioural similarity is not sufficient for observing similar neural dynamics.
I participated in three conferences to present and discuss my results at different stages of the project. Namely, I presented in FENS Forum 2022 (organised by the Federation of European Neuroscience Societies), NCM 2022 (organised by the Society for the Neural Control of Movement), and in iBAGS 2023 (organised by the International Basal Ganglia Society). This work was finally published as a pre-print in September 2022, and its final revised version is also published open access in Nature. The entire codebase to reproduce every figure and most of the datasets used in the paper are also available online.
The results stated above are for the full duration of the project. However, I will continue to work toward performing the experiments proposed in this project. The computational method developed and published during this project has the potential to both change the way BCIs are currently built, and accelerate their deployment in clinical settings. Indeed, using this method, one can readily use a BCI built on one user on another. Thus, our results could impact the wider society, including the large group of neurological patients who can benefit from BCI-based interventions.