Mikhail I. Krivonosov

Background/Objectives: Synolitic graphs (SGs) were developed for task-based fMRI, where edge weights encode the discriminative power of pairwise regional features; whether similar information can be recovered from resting-state data was untested. We benchmarked SGs for autism spectrum disorder (ASD) classification using the multisite ABIDE-I dataset (871 subjects: 403 subjects with ASD, 468 typical controls; 17 sites; CC200 atlas). Methods: Using 5-fold cross-validation with 10 repeats and balanced accuracy as the primary metric, we compared SGs with eight baselines including correlation matrices, tangent space connectivity, elastic net logistic regression, SVM, and XGBoost. Results: The vectorised correlation matrix achieved 67.9% balanced accuracy (AUC = 0.743), the combined model 68.0% (AUC = 0.744), and SGs 57.0% (AUC = 0.598). SGs performed above chance (permutation p = 0.001), but their performance was not significantly different from that of direct logistic regression (Nadeau–Bengio p = 0.82). After covariate residualisation, correlation-based methods retained 66.9–67.1% balanced accuracy, while SGs achieved 55.6%. Leave-one-site-out validation yielded 66.0–66.2% for correlation-based models and 55.0% for SGs, with similar declines of approximately 2 percentage points from standard cross-validation. SG-derived ROI rankings were unstable, and ADOS severity prediction among 193 ASD participants was underpowered and inconclusive. Conclusions: Conventional connectivity representations substantially outperformed SGs for resting-state ASD classification, contrasting with previously reported task-fMRI advantages. No evaluated method achieved performance suitable for stand-alone clinical screening or diagnosis; this study should therefore be interpreted as a methodological benchmark rather than a validation of a clinically deployable tool.

Background: Post-COVID-19 condition (PCC) affects many survivors, with evidence of sex-specific differences in prevalence and symptom profiles. However, few prediction studies have examined whether sex-stratified models improve prediction or generalize across sexes. This study aimed primarily to develop and compare sex-stratified machine learning models for PCC prediction using routinely available baseline variables, and secondarily to assess cross-sex generalizability and adversarial robustness. Methods: We analyzed a prospective longitudinal cohort of 1006 adults hospitalized with COVID-19 at Sechenov University Hospital Network (Moscow, Russia). Demographics, smoking status, and pre-existing comorbidities were extracted from medical records, and PCC status was assessed at 6-month follow-up. Machine learning models—including classical algorithms and graph-based neural networks—were trained separately for males and females. Cross-sex validation evaluated generalizability, variable importance aided interpretation, and adversarial perturbations assessed model robustness. Results: PCC prevalence was higher in females (53.9%) than males (39.1%). Overall predictive performance was modest across all models, with AUC values ranging approximately 0.50–0.61. Graph-based models achieved the highest discrimination, with the best AUC reaching approximately 0.61, while classical approaches provided limited predictive value. Cross-sex validation showed minor asymmetry: models trained on male data performed slightly better on female cases than vice versa. Adversarial testing revealed sensitivity of all models to input perturbations. Conclusions: Demographics and comorbidities alone provide insufficient information for reliable PCC prediction. Modest sex-specific differences in model generalizability suggest distinct, sex-associated PCC phenotypes, but richer multimodal data—including clinical biomarkers, wearable-derived measures, and patient-reported outcomes—will be required to develop clinically useful and equitable predictive models. Sex-stratified approaches should be considered in future post-viral syndrome prediction studies.

Background: Prostate cancer (PCa) is one of the most prevalent malignancies in men, and active surveillance (AS) is the recommended management strategy for low- and favourable intermediate-risk disease. Predicting which patients will progress during AS remains a clinical challenge. MRI-derived radiomics has shown promise for risk stratification, but conventional machine learning approaches treat radiomic features as independent variables and may not capture the complex inter-feature dependencies within imaging data. This study evaluates the application of Synolitic Graph Neural Networks (SGNNs) to baseline MRI-derived radiomic features for predicting prostate cancer progression on active surveillance. Methods: We studied 343 AS patients (73 progressors, 270 non-progressors) from a single-centre cohort prospectively enrolled between 2013 and 2019 and retrospectively analysed. Seventy-two radiomic features were extracted from baseline 3T MRI (T2-weighted imaging and apparent diffusion coefficient maps), together with three clinical variables (prostate volume, PSA, PSA density). The SGNN pipeline transformed each patient’s feature profile into a weighted graph encoding pairwise feature relationships via logistic regression classifiers trained within each cross-validation fold. GCN and GATv2 architectures were evaluated with multiple sparsification strategies and compared against Gradient Boosting, SVM, Random Forest, and logistic regression using 5-fold stratified cross-validation. Results: Among conventional methods, Gradient Boosting achieved the highest ROC-AUC (0.634 ± 0.080). The SGNN pipeline with GATv2, confidence-based sparsification (p = 0.8), and extended node features incorporating graph centrality measures achieved the best performance (ROC-AUC = 0.699 ± 0.044), an absolute improvement of 0.065 over the best conventional method. The addition of topological node features consistently improved performance by 3–5% across configurations. GATv2 outperformed GCN in matched settings. Conclusions: As a proof of concept, the SGNN framework achieved the highest mean ROC-AUC among the evaluated single-timepoint approaches, though results require confirmation in independent external cohorts. By encoding inter-feature relationships as patient-specific graphs, SGNN offers a complementary modelling paradigm for radiomic data in clinical oncology. Future work should incorporate longitudinal data and graph explainability methods.

Biological clocks aim to estimate the physiological state of an organism, or of specific organs or systems, from the perspective of aging. The inferred individual age acceleration is presumably linked to increased mortality and morbidity risks. Here, we introduce a cognitive aging clock based on a battery of five tests (sensomotor arithmetic, sensomotor mirror letter, 1-back, campimetry and the Stroop), capturing sensory and verbal perception, working memory, and decision-making in the context of cognitive conflict. The developed LightGBM machine learning model has the cognitive age MAE of 7.6 years. SHAP analysis yields the overall and individualized interpretations of accelerated aging. In particular, we demonstrate the dominant contribution of Stroop and campimetry test variables to cognitive age. The cognitive age is accelerated in the patients with dementia, whereas moderate cognitive impairment also shows the acceleration trend, but lacks statistical significance. MoCA test score is negatively associated with cognitive age deviation.

Background/Objectives: Gliomas are complex and heterogeneous brain tumors characterized by an unfavorable clinical course and a fatal prognosis, which can be improved by an early determination of tumor kind. Here, we developed explainable machine learning (ML) models for classifying three major glioma subtypes (astrocytoma, oligodendroglioma, and glioblastoma) and predicting survival rates based on RNA-seq data. Methods: We analyzed publicly available datasets and applied feature selection techniques to identify key biomarkers. Using various ML models, we performed classification and survival analysis to develop robust predictive models. The best-performing models were then interpreted using Shapley additive explanations (SHAP). Results: Thirteen key genes (TERT, NOX4, MMP9, TRIM67, ZDHHC18, HDAC1, TUBB6, ADM, NOG, CHEK2, KCNJ11, KCNIP2, and VEGFA) proved to be closely associated with glioma subtypes as well as survival. Support Vector Machine (SVM) turned out to be the optimal classification model with the balanced accuracy of 0.816 and the area under the receiver operating characteristic curve (AUC) of 0.896 for the test datasets. The Case-Control Cox regression model (CoxCC) proved best for predicting survival with the Harrell’s C-index of 0.809 and 0.8 for the test datasets. Using SHAP we revealed the gene expression influence on the outputs of both models, thus enhancing the transparency of the prediction generation process. Conclusions: The results indicated that the developed models could serve as a valuable practical tool for clinicians, assisting them in diagnosing and determining optimal treatment strategies for patients with glioma.

This paper introduces a novel approach for classifying multidimensional physiological and clinical data using Synolitic Graph Neural Networks (SGNNs). SGNNs are particularly good for addressing the challenges posed by high-dimensional datasets, particularly in healthcare, where traditional machine learning and Artificial Intelligence methods often struggle to find global optima due to the “curse of dimensionality”. To apply Geometric Deep Learning we propose a synolitic or ensemble graph representation of the data, a universal method that transforms any multidimensional dataset into a network, utilising only class labels from training data. The paper demonstrates the effectiveness of this approach through two classification tasks: synthetic and fMRI data from cognitive tasks. Convolutional Graph Neural Network architecture is then applied, and the results are compared with established machine learning algorithms. The findings highlight the robustness and interpretability of SGNNs in solving complex, high-dimensional classification problems.

Background: Respiratory viral infections (RVIs) exhibit seasonal patterns influenced by biological, ecological, and climatic factors. Weather variables such as temperature, humidity, and wind impact the transmission of droplet-borne viruses, potentially affecting disease severity. However, the role of climate in predicting complications in pediatric RVIs remains unclear, particularly in the context of climate-change-driven extreme weather events. Methods: This retrospective cohort study analyzed 1610 hospitalization records of children (0–18 years) with lower respiratory tract infections in Rome, Italy, between 2018 and 2023. Viral pathogens were identified using nasopharyngeal molecular testing, and weather data from the week preceding hospitalization were collected. Several machine learning models were tested, including logistic regression and random forest, comparing the baseline (demographic and clinical) models with those including climate variables. Results: Logistic regression showed a slight improvement in predicting severe RVIs with the inclusion of weather variables, with accuracy increasing from 0.785 to 0.793. Average temperature, dew point, and humidity emerged as significant contributors. Other algorithms did not demonstrate similar improvements. Conclusions: Climate variables can enhance logistic regression models’ ability to predict RVI severity, but their inconsistent impact across algorithms highlights challenges in integrating environmental data into clinical predictions. Further research is needed to refine these models for use in reliable healthcare applications.

Calcium imaging of nerve activity in the mouse hippocampus provides insights into cell-to-cell interactions [1]. Over time, fluctuations in cellular calcium levels encode distinct brain states during mouse brain imaging. Although synapses connect cells and transmit information, it is challenging to detect the spatiotemporal transmission pathway via calcium imaging of multiple cells due to the complexity of the spatial structure and imaging in a separate focal plane [2]. Reconstruction of the dynamic graph of connections between cells was proposed to overcome this problem. A dynamic graph comprises of individual graphs for each time point. A graph linked to a particular time t is composed of vertices representing cells and edges that depict the transmission of signals between cells at that moment. This paper puts forth 3 approaches for creating networks. The first method considers the overlapping time intervals of calcium events in individual cells [3]. The link is established between the cell with the earlier event and the cell with the later start of the event during the moments when the events occurred simultaneously in separate cells. Alternatively, time intervals between the start of events were taken into account. Potentially connected events that began no later than 2 s were linked, with the edge drawn from the cell depicting the earlier starting event to the second cell. The third technique involved linking sequentially occurring events in distinct cells. The edge is drawn from all active cells at the previous time step to newly activated cells at the subsequent time step. Data analysis is based on an experiment in which a mouse moved along a circular track while the fluorescence of neuronal calcium activity and the mouse’s position were recorded at a frequency of 20 frames per second [4]. A red dot was marked on the mouse’s head to track its position. The reconstructed dynamic graph was compared to the angular coordinate of the mouse on the ring to look for repeating patterns of activity. The racetrack was divided into 20 overlapping sectors, each spanning 36 degrees. The reconstructed networks were then assigned to sectors that aligned with the mouse’s angular position on the track. Next, we estimated the frequency of individual edge repetitions within each network group. Only edges that occurred at least three times were chosen for further analysis. We found repeated activations of various cell pairs that corresponded to clockwise and counterclockwise movement within the same sector. Furthermore, we identified the presence of alternating activation, where activity occurred in the first cell, then the second cell, and then again in the first cell. In addition, we identified complex sequences of 5–6 non-sequential activations, represented as a digraph without cycles, which is typical for single sectors.

Mammalian brains operate in very special surroundings: to survive they have to react quickly and effectively to the pool of stimuli patterns previously recognized as danger. Many learning tasks often encountered by living organisms involve a specific set-up centered around a relatively small set of patterns presented in a particular environment. For example, at a party, people recognize friends immediately, without deep analysis, just by seeing a fragment of their clothes. This set-up with reduced “ontology” is referred to as a “situation.” Situations are usually local in space and time. In this work, we propose that neuron-astrocyte networks provide a network topology that is effectively adapted to accommodate situation-based memory. In order to illustrate this, we numerically simulate and analyze a well-established model of a neuron-astrocyte network, which is subjected to stimuli conforming to the situation-driven environment. Three pools of stimuli patterns are considered: external patterns, patterns from the situation associative pool regularly presented to the network and learned by the network, and patterns already learned and remembered by astrocytes. Patterns from the external world are added to and removed from the associative pool. Then, we show that astrocytes are structurally necessary for an effective function in such a learning and testing set-up. To demonstrate this we present a novel neuromorphic computational model for short-term memory implemented by a two-net spiking neural-astrocytic network. Our results show that such a system tested on synthesized data with selective astrocyte-induced modulation of neuronal activity provides an enhancement of retrieval quality in comparison to standard spiking neural networks trained via Hebbian plasticity only. We argue that the proposed set-up may offer a new way to analyze, model, and understand neuromorphic artificial intelligence systems.