Mosquitos are the main transmissive agents of arboviral diseases. Manual classification of their neuronal spike patterns is very labor-intensive and expensive. Most available deep learning solutions require fully labeled spike datasets and highly preprocessed neuronal signals. This reduces the feasibility of mass adoption in actual field scenarios. To address the scarcity of labeled data problems, we propose a new Generative Adversarial Network (GAN) architecture that we call the Semi-supervised Swin-Inspired GAN (SSI-GAN). The Swin-inspired, shifted-window discriminator, together with a transformer-based generator, is used to classify neuronal spike trains and, consequently, detect viral neurotropism. We use a multi-head self-attention model in a flat, window-based transformer discriminator that learns to capture sparser high-frequency spike features. Using just 1 to 3% labeled data, SSI-GAN was trained with more than 15 million spike samples collected at five-time post-infection and recording classification into Zika-infected, dengue-infected, or uninfected categories. Hyperparameters were optimized using the Bayesian Optuna framework, and performance for robustness was validated under fivefold Monte Carlo cross-validation. SSI-GAN reached 99.93% classification accuracy on the third day post-infection with only 3% labeled data. It maintained high accuracy across all stages of infection with just 1% supervision. This shows a 97-99% reduction in manual labeling effort relative to standard supervised approaches at the same performance level. The shifted-window transformer design proposed here beat all baselines by a wide margin and set new best marks in spike-based neuronal infection classification.
Zika virus (ZIKV) and Dengue virus (DENV) belong to the Flaviviridae family and the Ortho flavivirus genus [1]. Other viruses of this family include West Nile (WNV), Japanese encephalitis (JEV), and Yellow fever virus (YFV) [1]. ZIKV was identified in a rhesus macaque in Uganda in 1947 [2]. It has had outbreaks in Africa, the Americas, Asia, and the Pacific region since 2007 and has become a serious global health concern over time [3]. This virus was recorded in more than 50 countries and infected around 390 million people worldwide as of now [4]. During these times, serious nervous system disorders, such as Guillain-Barré syndrome, encephalitis, myelitis, and neuralgia, and also developmental problems such as microcephaly and eye malformations, were largely observed [5].
ZIKV is mainly transmitted by Aedes aegypti mosquitoes [6]. It can also spread directly between humans through mother to fetus, sexual contact, and blood transfusion [7]. Studying how viruses affect mosquitos’ behavior and their nervous system, as a part of the virus life cycle, will help us to understand the transmission of mosquito-borne diseases [8]. There are many evolutionary similarities in the nervous systems of mammals and insects [9]. Both of them use the same chemical messengers and receptors [10]. The similar basic structure of the receptors in both groups allows us to compare functions in their bodies [11]. Their nervous systems also have structural and functional similarities [12]. There is a central nervous system and a peripheral nervous system in both species [13]. There are some differences, but these similarities suggest that studying ZIKV and DENV neural activity in insects can help understand their pathogenesis and infection mechanisms in humans [14].
ZIKV’s effects on the nervous system have been widely studied using different models, including in vitro studies with human pluripotent stem cell (hPSC)-derived neural progenitor cells and organoids, as well as in-vivo research in mouse models by means of different viral strains [15]. In a previous study, which focused on the mosquito life cycle, neural spikes produced by uninfected, ZIKV-infected, and DENV-infected Aedes aegypti were recorded by microelectrode array technology [16]. Analyzing the behavior and neural spikes showed virus neurotropism, which affects mosquito neurons and causes an increase in neural activity and behavioral changes [16]. These changes included different flight movements, more bites, more feeding, and effects on reproductive performance, such as fecundity and fertility [16]. These changes can make viral transmission easier by increasing mosquito-human interactions [16].
Neural spike classification is effective for analyzing the behavior of the neurological system among different methods, because it isolates spikes, the brain’s signals for transmitting information [17]. Studying mosquito neuronal activity and neurons’ spike classification is important for understanding the viral infection mechanisms of ZIKV and DENV [18]. Using deep learning to classify spikes in these insects can help us understand how these viruses affect both mosquitoes and humans [19]. Deep learning can make data processing automated and improve accuracy by reducing human error in classification tasks [20]. In addition, it can identify patterns and correlations in neuronal activity that may not be easily recognizable through traditional methods [21]. There are almost no studies that have focused on deep learning-based spike classification of insect data for ZIKV and DENV detection [18] using semi-supervised learning.
Review shows that several traditional methods have been used for spike classification [22]. For example, Lewicki used a Bayesian approach [23], while Letelier and Weber used wavelet analysis to classify neural spikes based on their time-frequency features, isolating signals from noise and overlapping activity [24]. Haas, Cohen et al. proposed a variance-based template matching model, which uses a CMOS variance estimator to classify data in real time [25]. Takekawa, Isomura et al. introduced a hybrid algorithm that combines multimodality-weighted principal component analysis and clustering techniques [26]. Traditional models perform well on certain benchmarks, but often exhibit low accuracy and poor generalization, and most of the time, they work effectively only on specific datasets [27]. On the other hand, deep learning models show impressive performance according to their ability to filter noise, enhance signal decoding, and produce more reliable results [28]. For example, Meyer, Schanze, et al. [28] developed a single-layer multilayer perceptron network of spike waveforms. Their results show that the model has a good performance, with accuracy between 89%-95% depending on the noise level [29]. Liu, Feng et al. proposed a method that uses convolutional neural networks (CNNs) and long short-term memory for spike classification [30]. Li, Tang et al. proposed
This content is AI-processed based on open access ArXiv data.