Jian. K. Liu

Neural circuit research requires resolving the functional coding mechanisms at the single-neuron scale. However, neural signals captured from extracellular recordings are inherently complex, shaped both by electrode properties and the spatially distributed neural population activity across recording sites. These factors introduce strong signal mixing and waveform variation, making it difficult for existing spike sorting methods to reliably isolate individual neurons from large-scale recordings. To address this, we develop NeuroSort, which fuses spatiotemporal information through a contrastive learning strategy to adaptively model neural dynamics. NeuroSort is a GPU-enabled, highly efficient spike sorting framework, enabling the isolation of up to thousands of neurons. We validated NeuroSort on in vivo extracellular recordings, including Utah array data from rhesus macaques performing motor tasks and Neuropixels 1.0 probe data from a rat during rest. Moreover, in a recording collected by the 1,024-channel Neuroscroll probe, NeuroSort sorted the largest number of well-isolated neurons from different brain areas while achieving a 2-8 times speedup over other methods. Taken together, NeuroSort generalizes across recording conditions, providing an interpretable and scalable paradigm for whole-brain dynamic analysis.

The accurate prediction of protein-ligand binding affinity is important for drug discovery yet remains challenging for multi-domain proteins, where inter-domain dynamics and flexible linkers govern molecular recognition. Current geometric deep learning methods typically treat proteins as monolithic graphs, failing to capture the distinct geometric and energetic signals at domain interfaces. To address this, we introduce DAGML (Domain-Aware Geometric Multimodal Learning), a hierarchical framework that explicitly models domain modularity. DAGML integrates a pre-trained protein language model with a novel domain-aware geometric encoder to distinguish intra- and inter-domain features, while a motif-centric ligand encoder captures pharmacophoric compatibility. We further curate a specialized multi-domain affinity benchmark, classifying complexes by binding topology (e.g., interface vs linker binders). Extensive experiments demonstrate that DAGML achieves a 21% reduction in MSE and a Pearson correlation of 0.726 compared to strong baselines. Ablation studies reveal that explicit modeling of domain interfaces is the primary driver of this improvement, particularly for ligands binding in the clefts between structural units. The code is available at https://github.com/jiankliu/DAGML.

Seizures are frequent complications in brain tumor patients, yet the underlying neuronal mechanisms remain poorly defined. Here, we examined pathophysiological alterations in the peritumoral cortex of patients undergoing tumor resection. The synaptic activity, dendritic spine density, and gene expression of peritumoral pyramidal neurons differed significantly between patients with and without seizures. Using an inducible glioma rodent model, we characterized the progression of these alterations and their predictive value for seizure initiation. Computational simulations revealed that human cortical neurons are highly susceptible to synaptic and dendritic perturbations, which induce paroxysmal depolarizing shifts (PDS) in affected networks. Longitudinal analyses post-surgery showed that PDS were detectable prior to seizure onset in a subset of patients and reliably predicted post-resection seizure occurrence. These findings elucidate key neuronal substrates of tumor-associated seizures and suggest PDS as a potential biomarker for seizure risk, offering a foundation for targeted diagnostic and therapeutic strategies.

Protein language models (PLMs) have demonstrated remarkable capabilities in learning relationships between protein sequences and functions. However, finetuning these large models requires substantial computational resources, often with suboptimal task-specific results. This study investigates how parameter-efficient finetuning via LoRA can enhance protein property prediction while significantly reducing computational demands. By applying LoRA to ESM-2 and ESM-C models of varying sizes and evaluating 10 diverse protein property prediction tasks, we demonstrate that smaller models with LoRA adaptation can match or exceed the performance of larger models without adaptation. Additionally, we integrate contact map information through a multi-head attention mechanism, improving model comprehension of structural features. Our systematic analysis reveals that LoRA finetuning enables faster convergence, better performance, and more efficient resource utilization, providing practical guidance for protein research applications in resource-constrained environments. The code is available at https://github.com/jiankliu/SeqProFT.

The success of deep learning methods over the past decade has been partially shrouded in the shadow of adversarial attacks. Even a tiny undetectable deformation can lead to vicious misleading targeted at safety-critical applications. In contrast, the brain is far more robust when performing complex cognitive tasks. Nevertheless, the underlying mechanisms that contribute to the brain’s high reliability remain largely unexplored. At the intersection of neuroscience and artificial intelligence, we show that neuromorphic paradigms offer a promising solution to the dilemma brought by deep learning’s inherent vulnerabilities. Specifically, we exploit the temporal processing capabilities of spiking neural networks (SNNs) to achieve robustness surpassing that of traditional artificial neural networks (ANNs). We demonstrate that prioritizing task-critical information in the encoded sequence and employing early exit decoding to ignore later perturbations significantly enhance SNN robustness. Further improvements in robustness are achieved by accurately capturing temporal dependencies through specialized training algorithms. Additionally, we introduce a fusion encoding strategy to balance SNN generalization on natural data with robustness against adversarial input. Experimental results on the CIFAR-10 dataset show that SNNs trained with these combined methods achieve twice the robustness of ANNs. Overall, our work demonstrates that neuromorphic computing, leveraging the temporal processing capabilities of SNNs, not only provides superior robustness compared to ANNs but also retains the benefit of low energy consumption. These advancements pave the way for developing next-generation, environmentally friendly, and reliable spike-based intelligent systems.

Spiking neural networks (SNNs) have garnered interest due to their energy efficiency and superior effectiveness on neuromorphic chips compared with traditional artificial neural networks (ANNs). One of the mainstream approaches to implementing deep SNNs is the ANN–SNN conversion, which integrates the efficient training strategy of ANNs with the energy-saving potential and fast inference capability of SNNs. However, under extremely low-latency conditions, the existing conversion theory suggests that the problem of SNNs’ neurons firing more or fewer spikes within each layer, i.e., residual error, leads to a performance gap in the converted SNNs compared with the original ANNs. This severely limits the possibility of the practical application of SNNs on delay-sensitive edge devices. Existing conversion methods addressing this problem usually involve modifying the state of the conversion spiking neurons. However, these methods do not consider their adaptability and compatibility with neuromorphic chips. We propose a new approach based on explicit modeling of residual errors as additive noise. The noise is incorporated into the activation function of the source ANN, effectively reducing the impact of residual error on SNN performance. Our experiments on the CIFAR10/100 and Tiny-ImageNet datasets verify that our approach exceeds the prevailing ANN–SNN conversion methods and directly trained SNNs concerning accuracy and the required time steps. Overall, our method provides new ideas for improving SNN performance under ultralow-latency conditions and is expected to promote practical neuromorphic hardware applications for further development. The code for our NQ framework is available at https://github.com/hzp2022/ANN2SNN_NQ