Annamaria Defilippo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Annamaria Defilippo has a total of 40 co-authors. Here are the top co-authors by number of shared publications:
- Pietro Hiram Guzzi
Department of Surgical and Medical Sciences, Magna Graecia University of Catanzaro, viale Europa, 88100 Catanzaro, Italy
15 shared publications - Pierangelo Veltri
Department of Computer Engineering, Modeling, Electronics and Systems (DIMES), University of Calabria, via P. Bucci, 87036 Rende, Italy
14 shared publications - Ugo Lomoio
Department of Surgical and Medical Sciences, Magna Graecia University of Catanzaro, viale Europa, 88100 Catanzaro, Italy
9 shared publications - Barbara Puccio
Magna Graecia University of Catanzaro
6 shared publications - Fabio Scarpa
University of Sassari
6 shared publications - Francesco Branda
Unit of Medical Statistics and Molecular Epidemiology Università Campus Bio‐Medico di Roma Rome Italy
6 shared publications
Latest Publications
- Open Access
- 26 October 2026
- Published by MDPI AG in Tropical Medicine and Infectious Disease
- Vol. 11 (11) , 307
MalarIA: A Modular AI-Powered Web Platform for Automated Plasmodium Species Classification from Blood Smear Microscopy
Background: Malaria continues to pose a significant global public health challenge, as reflected in the estimated 282 million cases and 610,000 associated deaths recorded worldwide in 2024. Elimination efforts are threatened by antimalarial drug resistance, relapsing Plasmodium vivax infections, and emerging zoonotic Plasmodium knowlesi. Accurate species-level identification is critical for appropriate treatment and surveillance, yet conventional microscopy is labor-intensive and operator-dependent, while rapid diagnostic tests lack sensitivity and species-discrimination capacity. This diagnostic gap underscores the need for automated, trustworthy AI-based tools. Methods: We present MalarIA, a modular, extensible AI-based web platform designed to translate deep learning research into deployable malaria diagnostics. The platform integrates three architectures, ResNet50, Vision Transformer (ViT), and a hybrid ResNetViT, that our group previously trained from scratch on 1038 real-world blood smear images spanning four Plasmodium species (P. falciparum, P. vivax, P. malariae, and P. ovale). MalarIA features a model registry for seamless integration of novel architectures, five Class Activation Mapping (CAM) variants for explainability, Expected Calibration Error (ECE) for confidence reporting, and a validation module (MalarIA Validator) supporting human-in-the-loop active learning and multicenter external validation. Results: The contribution of this article is the design and implementation of the platform itself; the predictive performance results summarized here are reused from our previous cross-validation study and no new model training is claimed in the present work. In that earlier evaluation, ResNet50 achieved the highest accuracy (96.9%, F1 96.3%, ROC-AUC 0.997, macro-averaged AUC-PR 0.990, Cohen’s kappa 0.939). The hybrid ResNetViT demonstrated superior calibration (ECE 0.0175 vs. 0.0263 for ResNet50 and 0.0778 for ViT, recomputed from the archived out-of-fold predictions) and enhanced sensitivity for minority species (P. malariae recall 0.970; P. vivax 0.978). The platform is demonstrated end-to-end through its web interface, from image upload to explainable species-level results. Conclusions: MalarIA demonstrates that translational AI, grounded in extensibility, explainability, rigorous calibration, and practical deployment, can effectively bridge the gap between research and clinical impact in malaria diagnostics. The platform is currently a research-grade prototype: the interface, model registry, explainability and calibration modules are implemented and demonstrated, while full backend model serving, multicenter validation, and regulatory assessment remain planned future developments.
- 26 October 2026
- Published by MDPI AG in Tropical Medicine and Infectious Disease
- Vol. 11 (11) , 307
Open AccessMalarIA: A Modular AI-Powered Web Platform for Automated Plasmodium Species Classification from Blood Smear Microscopy
Background: Malaria continues to pose a significant global public health challenge, as reflected in the estimated 282 million cases and 610,000 associated deaths recorded worldwide in 2024. Elimination efforts are threatened by antimalarial drug resistance, relapsing Plasmodium vivax infections, and emerging zoonotic Plasmodium knowlesi. Accurate species-level identification is critical for appropriate treatment and surveillance, yet conventional microscopy is labor-intensive and operator-dependent, while rapid diagnostic tests lack sensitivity and species-discrimination capacity. This diagnostic gap underscores the need for automated, trustworthy AI-based tools. Methods: We present MalarIA, a modular, extensible AI-based web platform designed to translate deep learning research into deployable malaria diagnostics. The platform integrates three architectures, ResNet50, Vision Transformer (ViT), and a hybrid ResNetViT, that our group previously trained from scratch on 1038 real-world blood smear images spanning four Plasmodium species (P. falciparum, P. vivax, P. malariae, and P. ovale). MalarIA features a model registry for seamless integration of novel architectures, five Class Activation Mapping (CAM) variants for explainability, Expected Calibration Error (ECE) for confidence reporting, and a validation module (MalarIA Validator) supporting human-in-the-loop active learning and multicenter external validation. Results: The contribution of this article is the design and implementation of the platform itself; the predictive performance results summarized here are reused from our previous cross-validation study and no new model training is claimed in the present work. In that earlier evaluation, ResNet50 achieved the highest accuracy (96.9%, F1 96.3%, ROC-AUC 0.997, macro-averaged AUC-PR 0.990, Cohen’s kappa 0.939). The hybrid ResNetViT demonstrated superior calibration (ECE 0.0175 vs. 0.0263 for ResNet50 and 0.0778 for ViT, recomputed from the archived out-of-fold predictions) and enhanced sensitivity for minority species (P. malariae recall 0.970; P. vivax 0.978). The platform is demonstrated end-to-end through its web interface, from image upload to explainable species-level results. Conclusions: MalarIA demonstrates that translational AI, grounded in extensibility, explainability, rigorous calibration, and practical deployment, can effectively bridge the gap between research and clinical impact in malaria diagnostics. The platform is currently a research-grade prototype: the interface, model registry, explainability and calibration modules are implemented and demonstrated, while full backend model serving, multicenter validation, and regulatory assessment remain planned future developments.
- Open Access
A Computational Workflow Integrating Explainable AI and Database Mining for the Prioritization of Antimicrobial Natural Products
The escalating threat of antimicrobial resistance demands the discovery of antibacterial agents with unprecedented chemical scaffolds. Natural products have historically been the most prolific source of antibiotics, yet the vast majority of catalogued natural products remain untested for antimicrobial activity. Here we present Natural-Product Graph–language, Uncertainty-Aware Relational Discovery (NP-GUARD), a modular computational framework that reframes antimicrobial natural-product discovery as an uncertainty-aware, relational, multi-objective learning problem. NP-GUARD integrates six core components: provenance-tracked data curation, multi-view molecular representation, pathogen-conditioned multi-task prediction, explicit uncertainty quantification, stability-aware explainability, and conformal-Pareto prioritization, all consolidated within a queryable knowledge base. Of these components, the present study implements and benchmarks the data-curation, predictive, uncertainty, explainability and prioritization layers, including a first pathogen-conditioned masked multi-task graph model, while the full multi-view encoder and the knowledge base are specified for subsequent work. We report a proof-of-concept evaluation on public screening data for Staphylococcus aureus (38,995 compounds), Escherichia coli (2291) and Pseudomonas aeruginosa (29,986), together with human cytotoxicity data and a 406,959-compound screening library from COCONUT. Under repeated scaffold-aware splitting, baseline models achieve strong discrimination (AUROC 0.78–0.93) but reveal substantial split-to-split variance in minority-class performance; class-conditional conformal prediction attains near-nominal per-class coverage; and feature attributions show only moderate stability across independently trained models (top-20 Jaccard 0.54), underscoring a critical caveat for explanation-driven discovery. Screening the COCONUT library validates the uncertainty-aware activity ranking, achieving an 87-fold enrichment of known antibiotics in the top 100 candidates (128-fold within the fully characterized named-compound stratum), although analogue-exclusion analysis shows this signal is dominated by rediscovery of training-like chemistry. The novelty-seeking Pareto layer yields a structurally diverse 90-compound shortlist spanning 72 scaffolds, deliberately steered toward unexplored chemotypes; we emphasize that the shortlist constitutes prioritized hypotheses for experimental testing, not validated actives. NP-GUARD provides a transparent, reproducible and decision-oriented foundation for prioritizing natural products for experimental validation.
Open AccessA Computational Workflow Integrating Explainable AI and Database Mining for the Prioritization of Antimicrobial Natural Products
The escalating threat of antimicrobial resistance demands the discovery of antibacterial agents with unprecedented chemical scaffolds. Natural products have historically been the most prolific source of antibiotics, yet the vast majority of catalogued natural products remain untested for antimicrobial activity. Here we present Natural-Product Graph–language, Uncertainty-Aware Relational Discovery (NP-GUARD), a modular computational framework that reframes antimicrobial natural-product discovery as an uncertainty-aware, relational, multi-objective learning problem. NP-GUARD integrates six core components: provenance-tracked data curation, multi-view molecular representation, pathogen-conditioned multi-task prediction, explicit uncertainty quantification, stability-aware explainability, and conformal-Pareto prioritization, all consolidated within a queryable knowledge base. Of these components, the present study implements and benchmarks the data-curation, predictive, uncertainty, explainability and prioritization layers, including a first pathogen-conditioned masked multi-task graph model, while the full multi-view encoder and the knowledge base are specified for subsequent work. We report a proof-of-concept evaluation on public screening data for Staphylococcus aureus (38,995 compounds), Escherichia coli (2291) and Pseudomonas aeruginosa (29,986), together with human cytotoxicity data and a 406,959-compound screening library from COCONUT. Under repeated scaffold-aware splitting, baseline models achieve strong discrimination (AUROC 0.78–0.93) but reveal substantial split-to-split variance in minority-class performance; class-conditional conformal prediction attains near-nominal per-class coverage; and feature attributions show only moderate stability across independently trained models (top-20 Jaccard 0.54), underscoring a critical caveat for explanation-driven discovery. Screening the COCONUT library validates the uncertainty-aware activity ranking, achieving an 87-fold enrichment of known antibiotics in the top 100 candidates (128-fold within the fully characterized named-compound stratum), although analogue-exclusion analysis shows this signal is dominated by rediscovery of training-like chemistry. The novelty-seeking Pareto layer yields a structurally diverse 90-compound shortlist spanning 72 scaffolds, deliberately steered toward unexplored chemotypes; we emphasize that the shortlist constitutes prioritized hypotheses for experimental testing, not validated actives. NP-GUARD provides a transparent, reproducible and decision-oriented foundation for prioritizing natural products for experimental validation.
- Open Access
- 27 July 2026
- Published by MDPI AG in Infectious Disease Reports
- Vol. 18 (4) , 78
Deep Learning-Based Malaria Parasite Image Classification on Real Microscopy Data
Background: Accurate and timely malaria diagnosis with species-level identification of Plasmodium parasites is critical for guiding effective treatment and disease management. Although light microscopy remains the diagnostic gold standard, its dependence on highly trained personnel limits accessibility, particularly in resource-constrained settings. Recent advances in deep learning have enabled automated image-based diagnosis with promising performance; however, reliable differentiation among Plasmodium species continues to pose a major challenge. This study aims to evaluate and compare the effectiveness of different deep learning architectures for automated species identification from microscopy images. Methods: Three deep learning architectures were systematically assessed: a convolutional backbone (ResNet50), a Vision Transformer (ViT), and a hybrid ResNetViT model. All models were trained from scratch, without using pretrained weights, on a dataset comprising real-world thick blood smear images augmented with publicly available microscopy data from Kaggle. In the hybrid architecture, the ResNet module was used to extract robust local morphological features, while the ViT component captured long-range dependencies and contextual relationships within images. Results: In cross-validation experiments, all three architectures consistently achieved high diagnostic performance. ResNet50 attained the highest accuracy (96.9%), an F1-score of 96.3%, and an ROC-AUC of 0.997. The Vision Transformer achieved 93.1% accuracy, 91.7% F1-score, and an ROC-AUC of 0.989. The hybrid ResNetViT reached 95.2% accuracy, 94.2% F1-score, and an ROC-AUC of 0.995. These results confirm that all architectures can reliably distinguish among Plasmodium species. Although ResNet50 achieved the highest raw accuracy, the hybrid model showed the most stable calibration and the closest qualitative agreement between its attention maps and expert-annotated parasite locations. Conclusions: The findings demonstrate that convolutional, transformer-based, and hybrid deep learning architectures can be successfully trained on real microscopy data for species-level malaria diagnosis. These results support the feasibility of deploying scalable, automated diagnostic systems to improve both accuracy and accessibility of malaria detection, particularly in resource-limited healthcare settings.
- 27 July 2026
- Published by MDPI AG in Infectious Disease Reports
- Vol. 18 (4) , 78
Open AccessDeep Learning-Based Malaria Parasite Image Classification on Real Microscopy Data
Background: Accurate and timely malaria diagnosis with species-level identification of Plasmodium parasites is critical for guiding effective treatment and disease management. Although light microscopy remains the diagnostic gold standard, its dependence on highly trained personnel limits accessibility, particularly in resource-constrained settings. Recent advances in deep learning have enabled automated image-based diagnosis with promising performance; however, reliable differentiation among Plasmodium species continues to pose a major challenge. This study aims to evaluate and compare the effectiveness of different deep learning architectures for automated species identification from microscopy images. Methods: Three deep learning architectures were systematically assessed: a convolutional backbone (ResNet50), a Vision Transformer (ViT), and a hybrid ResNetViT model. All models were trained from scratch, without using pretrained weights, on a dataset comprising real-world thick blood smear images augmented with publicly available microscopy data from Kaggle. In the hybrid architecture, the ResNet module was used to extract robust local morphological features, while the ViT component captured long-range dependencies and contextual relationships within images. Results: In cross-validation experiments, all three architectures consistently achieved high diagnostic performance. ResNet50 attained the highest accuracy (96.9%), an F1-score of 96.3%, and an ROC-AUC of 0.997. The Vision Transformer achieved 93.1% accuracy, 91.7% F1-score, and an ROC-AUC of 0.989. The hybrid ResNetViT reached 95.2% accuracy, 94.2% F1-score, and an ROC-AUC of 0.995. These results confirm that all architectures can reliably distinguish among Plasmodium species. Although ResNet50 achieved the highest raw accuracy, the hybrid model showed the most stable calibration and the closest qualitative agreement between its attention maps and expert-annotated parasite locations. Conclusions: The findings demonstrate that convolutional, transformer-based, and hybrid deep learning architectures can be successfully trained on real microscopy data for species-level malaria diagnosis. These results support the feasibility of deploying scalable, automated diagnostic systems to improve both accuracy and accessibility of malaria detection, particularly in resource-limited healthcare settings.
- Open Access
- 23 July 2026
- Published by MDPI AG in Preprints.org
Safety-by-Construction Agentic AI for Territorial Pre-Triage: A Proof-of-Concept Case Study for Rural and Inner Areas of Calabria
\textbf{Background:} Rural and inner areas combine geographic distance, demographic ageing, small municipalities, and fragmented access to health services. These conditions are especially relevant in Calabria, where a large proportion of municipalities have fewer than 5,000 residents and several officially designated inner areas show long travel times to reference service poles. Digital pre-triage may provide an additional access and orientation layer, but the use of generative artificial intelligence in health care raises concerns about hallucination, non-determinism, privacy, and under-triage.\textbf{Methods:} We designed a modular agentic-AI framework for territorial pre-triage according to a safety-by-construction principle. The architecture separates conversational elicitation (workflow W01) from clinical interpretation and priority assignment (workflow W02) through a pseudonymized JSON contract. W02 combines symptom normalization, red-flag detection, four function-specific retrieval-augmented generation collections, a panel of three role-diversified classifiers, and a deterministic reconciliation node based on majority or median voting with a hard safety override for confirmed red flags. The framework produces two outputs: practical care-pathway guidance for the citizen and a structured, auditable dossier for health professionals. Evaluation included 28 synthetic Italian-language cases processed directly by W02, nine end-to-end conversational cases, and 11 official Emergency Severity Index cases used as an external reference.\textbf{Results:} In the 28-case internal diagnostic test, exact agreement was 35.7\%, quadratic-weighted Cohen's $\kappa$ was 0.717, mean absolute error was 0.75 priority levels, and no under-triage occurred. Eighteen cases (64.3\%) were over-triaged, revealing an excessive conservative bias. Trace analysis identified four main failure mechanisms: missing objective measurements, overly conservative reconciliation, false red-flag activation from negated symptoms, and adult--paediatric retrieval contamination. Corrective changes preserved zero under-triage in the 11-case external evaluation, with reported quadratic-weighted $\kappa=0.77$, recall of 1.00 for the most critical class, and approximately 45\% exact agreement; only one discrepancy was judged a true retrieval error, while others reflected differences between Italian acuity-based coding and resource-oriented ESI criteria.\textbf{Conclusions:} The study supports the technical feasibility of a traceable agentic architecture for territorial pre-triage and provides an intervention blueprint tailored to rural and inner areas. It does not establish clinical effectiveness or safety. Prospective clinician-adjudicated validation, on-premise deployment, integration of vital signs, interoperability, human-factors testing, and regulatory assessment are required before real-world use.- 23 July 2026
- Published by MDPI AG in Preprints.org
Open AccessSafety-by-Construction Agentic AI for Territorial Pre-Triage: A Proof-of-Concept Case Study for Rural and Inner Areas of Calabria
\textbf{Background:} Rural and inner areas combine geographic distance, demographic ageing, small municipalities, and fragmented access to health services. These conditions are especially relevant in Calabria, where a large proportion of municipalities have fewer than 5,000 residents and several officially designated inner areas show long travel times to reference service poles. Digital pre-triage may provide an additional access and orientation layer, but the use of generative artificial intelligence in health care raises concerns about hallucination, non-determinism, privacy, and under-triage.\textbf{Methods:} We designed a modular agentic-AI framework for territorial pre-triage according to a safety-by-construction principle. The architecture separates conversational elicitation (workflow W01) from clinical interpretation and priority assignment (workflow W02) through a pseudonymized JSON contract. W02 combines symptom normalization, red-flag detection, four function-specific retrieval-augmented generation collections, a panel of three role-diversified classifiers, and a deterministic reconciliation node based on majority or median voting with a hard safety override for confirmed red flags. The framework produces two outputs: practical care-pathway guidance for the citizen and a structured, auditable dossier for health professionals. Evaluation included 28 synthetic Italian-language cases processed directly by W02, nine end-to-end conversational cases, and 11 official Emergency Severity Index cases used as an external reference.\textbf{Results:} In the 28-case internal diagnostic test, exact agreement was 35.7\%, quadratic-weighted Cohen's $\kappa$ was 0.717, mean absolute error was 0.75 priority levels, and no under-triage occurred. Eighteen cases (64.3\%) were over-triaged, revealing an excessive conservative bias. Trace analysis identified four main failure mechanisms: missing objective measurements, overly conservative reconciliation, false red-flag activation from negated symptoms, and adult--paediatric retrieval contamination. Corrective changes preserved zero under-triage in the 11-case external evaluation, with reported quadratic-weighted $\kappa=0.77$, recall of 1.00 for the most critical class, and approximately 45\% exact agreement; only one discrepancy was judged a true retrieval error, while others reflected differences between Italian acuity-based coding and resource-oriented ESI criteria.\textbf{Conclusions:} The study supports the technical feasibility of a traceable agentic architecture for territorial pre-triage and provides an intervention blueprint tailored to rural and inner areas. It does not establish clinical effectiveness or safety. Prospective clinician-adjudicated validation, on-premise deployment, integration of vital signs, interoperability, human-factors testing, and regulatory assessment are required before real-world use. - Open Access
- 30 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-2
Extracting Clinical Insights: LLM-Powered EHR Knowledge Graphs
Electronic Health Records (EHRs) represent a rich source of structured clinical information that is difficult to query at scale using traditional methods. This work proposes an end-to-end pipeline that parses synthetic EHR data, organises clinical entities into a Neo4j Knowledge Graph (KG), and enables LLM-powered question answering (QA) over patient data. The pipeline leverages open-source Qwen models for both embedding generation and natural language to Cypher query translation, combined with a LlamaIndex-based vector retrieval layer. A comparative evaluation between a vector-only RAG baseline and a KG-augmented RAG pipeline is performed using an LLM-as-Judge framework, with qualitative results suggesting improved reasoning consistency when structured graph knowledge is incorporated. An interactive Streamlit demo allows step-by-step exploration of the full pipeline.
- 30 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-2
Open AccessExtracting Clinical Insights: LLM-Powered EHR Knowledge Graphs
Electronic Health Records (EHRs) represent a rich source of structured clinical information that is difficult to query at scale using traditional methods. This work proposes an end-to-end pipeline that parses synthetic EHR data, organises clinical entities into a Neo4j Knowledge Graph (KG), and enables LLM-powered question answering (QA) over patient data. The pipeline leverages open-source Qwen models for both embedding generation and natural language to Cypher query translation, combined with a LlamaIndex-based vector retrieval layer. A comparative evaluation between a vector-only RAG baseline and a KG-augmented RAG pipeline is performed using an LLM-as-Judge framework, with qualitative results suggesting improved reasoning consistency when structured graph knowledge is incorporated. An interactive Streamlit demo allows step-by-step exploration of the full pipeline.
- Open Access
- 15 June 2026
- Published by MDPI AG in Preprints.org
Graph-Based Epidemic Intelligence: Explainable Neural Models for Pandemic Control and Zoonotic Transmission
Epidemics spread through contact, movement, behavior, and public health interventions, an inherently relational dynamic in which the infection travels along connections between people, places, or animal species. For this reason, we need mathematical and computational models capable of explicitly representing these connections. This paper introduces the theoretical foundations of network-based epidemic models, such as SIR and SEIR, and demonstrates how graph neural networks (GNNs) can learn the spatiotemporal patterns of transmission from data, overcoming the limitations of classical models. Three case studies are presented: measles, i.e., uneven vaccination coverage, COVID-19, i.e., targeted vaccination of the most central nodes in the contact network, and hantavirus, i.e., a multilevel model linking rodents, the environment, the molecular response, and human-to-human transmission). Since public health decisions must be justifiable, the work devotes particular attention to the explainability of the models: identifying which individuals, contacts, or territories are most critical and which alternative interventions could change the outcome of an epidemic. Finally, an operational pipeline is outlined to translate complex data into reliable and transparent decision support.
- 15 June 2026
- Published by MDPI AG in Preprints.org
Open AccessGraph-Based Epidemic Intelligence: Explainable Neural Models for Pandemic Control and Zoonotic Transmission
Epidemics spread through contact, movement, behavior, and public health interventions, an inherently relational dynamic in which the infection travels along connections between people, places, or animal species. For this reason, we need mathematical and computational models capable of explicitly representing these connections. This paper introduces the theoretical foundations of network-based epidemic models, such as SIR and SEIR, and demonstrates how graph neural networks (GNNs) can learn the spatiotemporal patterns of transmission from data, overcoming the limitations of classical models. Three case studies are presented: measles, i.e., uneven vaccination coverage, COVID-19, i.e., targeted vaccination of the most central nodes in the contact network, and hantavirus, i.e., a multilevel model linking rodents, the environment, the molecular response, and human-to-human transmission). Since public health decisions must be justifiable, the work devotes particular attention to the explainability of the models: identifying which individuals, contacts, or territories are most critical and which alternative interventions could change the outcome of an epidemic. Finally, an operational pipeline is outlined to translate complex data into reliable and transparent decision support.
- Open Access
- 8 June 2026
- Published by MDPI AG in Preprints.org
Biologically Informed Graph Contrastive Learning for Microbiome Data Analysis: A Survey
The human microbiome is a complex, dynamic and highly structured ecosystem whose analysis requires computational methods able to capture relationships among microbial taxa, genes, metabolic pathways, host factors, environmental exposures and disease phenotypes. Conventional machine-learning pipelines often represent microbiome samples as independent high-dimensional abundance vectors, thereby neglecting ecological, phylogenetic and functional dependencies among microbial entities. Graph-based learning provides a natural framework for modelling such dependencies, whereas graph contrastive learning (GCL) offers a self- supervised paradigm for learning robust representations from graph-structured data under limited label availability. This survey reviews the emerging intersection between GCL and microbiome data analysis. We first discuss the biological and computational characteristics of microbiome data, including sparsity, zero inflation, compositionality, batch effects, cohort heterogeneity and weak supervision. We then organize microbiome graph representations into taxa–taxa association networks, phylogenetic graphs, sample similarity graphs, microbe– disease association networks, host–microbe graphs, metabolic graphs and heterogeneous multi-omics graphs. Next, we summarize the foundations of GCL, including view generation, positive and negative pair construction, contrastive objectives, negative-free learning and multi-view representation learning.
- 8 June 2026
- Published by MDPI AG in Preprints.org
Open AccessBiologically Informed Graph Contrastive Learning for Microbiome Data Analysis: A Survey
The human microbiome is a complex, dynamic and highly structured ecosystem whose analysis requires computational methods able to capture relationships among microbial taxa, genes, metabolic pathways, host factors, environmental exposures and disease phenotypes. Conventional machine-learning pipelines often represent microbiome samples as independent high-dimensional abundance vectors, thereby neglecting ecological, phylogenetic and functional dependencies among microbial entities. Graph-based learning provides a natural framework for modelling such dependencies, whereas graph contrastive learning (GCL) offers a self- supervised paradigm for learning robust representations from graph-structured data under limited label availability. This survey reviews the emerging intersection between GCL and microbiome data analysis. We first discuss the biological and computational characteristics of microbiome data, including sparsity, zero inflation, compositionality, batch effects, cohort heterogeneity and weak supervision. We then organize microbiome graph representations into taxa–taxa association networks, phylogenetic graphs, sample similarity graphs, microbe– disease association networks, host–microbe graphs, metabolic graphs and heterogeneous multi-omics graphs. Next, we summarize the foundations of GCL, including view generation, positive and negative pair construction, contrastive objectives, negative-free learning and multi-view representation learning.
- Open Access
- 18 May 2026
- Published by MDPI AG in Preprints.org
Local Alignment of Differential Causal Networks: A Methodological Framework for Detecting Recurrent Rewiring Motifs Across Multiple Systems
Differential Causal Networks (DCNs) were introduced to represent changes between two causal networks inferred under different conditions. In their original use, however, DCNs remain pairwise objects: each differential graph summarizes rewiring within a single system, while common differential structures shared across many systems remain implicit. We introduce a methodological framework for the local alignment of DCNs aimed at detecting recurrent rewiring motifs, that is, small directed differential subnetworks that reappear across multiple systems under the same contrast. The proposed framework transforms each system-specific comparison into a signed directed differential graph, preserves both edge direction and change type, and searches for approximate local correspondences rather than enforcing a full-network mapping. The method consists of four steps: construction of signed DCNs, extraction of differential seeds, pairwise local alignment by seed-and-extend, and progressive multiple alignment to build consensus motifs. We define a score that combines node compatibility, differential-edge conservation, directional consistency, and recurrence support, and we complement the alignment procedure with null-model testing and robustness analysis. The result is a collection of consensus local differential modules ranked by recurrence, confidence, and statistical significance. In this formulation, DCNs become comparable units in a higher-order analysis whose goal is not merely to describe pairwise causal change, but to identify the same local rewiring logic reused across multiple systems.
- 18 May 2026
- Published by MDPI AG in Preprints.org
Open AccessLocal Alignment of Differential Causal Networks: A Methodological Framework for Detecting Recurrent Rewiring Motifs Across Multiple Systems
Differential Causal Networks (DCNs) were introduced to represent changes between two causal networks inferred under different conditions. In their original use, however, DCNs remain pairwise objects: each differential graph summarizes rewiring within a single system, while common differential structures shared across many systems remain implicit. We introduce a methodological framework for the local alignment of DCNs aimed at detecting recurrent rewiring motifs, that is, small directed differential subnetworks that reappear across multiple systems under the same contrast. The proposed framework transforms each system-specific comparison into a signed directed differential graph, preserves both edge direction and change type, and searches for approximate local correspondences rather than enforcing a full-network mapping. The method consists of four steps: construction of signed DCNs, extraction of differential seeds, pairwise local alignment by seed-and-extend, and progressive multiple alignment to build consensus motifs. We define a score that combines node compatibility, differential-edge conservation, directional consistency, and recurrence support, and we complement the alignment procedure with null-model testing and robustness analysis. The result is a collection of consensus local differential modules ranked by recurrence, confidence, and statistical significance. In this formulation, DCNs become comparable units in a higher-order analysis whose goal is not merely to describe pairwise causal change, but to identify the same local rewiring logic reused across multiple systems.
- Open Access
- 20 February 2026
- Published by Springer Nature in BMC Medical Informatics and Decision Making
- Vol. 26 (1)
From data to decisions: a modular platform for modelling and simulation of infectious disease diffusion in networks
Accurately modelling diffusion dynamics in complex networks is essential for improving medical outcomes, guiding pandemic preparedness, and optimizing reso
- 20 February 2026
- Published by Springer Nature in BMC Medical Informatics and Decision Making
- Vol. 26 (1)
Open AccessFrom data to decisions: a modular platform for modelling and simulation of infectious disease diffusion in networks
Accurately modelling diffusion dynamics in complex networks is essential for improving medical outcomes, guiding pandemic preparedness, and optimizing reso
- Open Access
- 2 February 2026
- Published by MDPI AG in Preprints.org
Real-Time Remote Patient Monitoring in Mountainous Regions: Architecture and Pilot Deployment of a Telemedicine Ecosystem
The rugged morphology and dispersed settlements of the Calabrian region pose long-standing barriers to timely and equitable access to healthcare, particularly for elderly and fragile populations living in mountainous areas. In this paper, we present a telemedicine ecosystem specifically designed for Calabria that integrates certified wearable wristbands, secure communication infrastructure, and intelligent back-end services to enable continuous home monitoring and rapid clinical intervention. Building on the SidlyCare platform, the system acquires real-time physiological signals (heart rate and oxygen saturation), streams them to a central server through encrypted channels, and applies machine learning and deep learning-based anomaly detection models to identify both acute and insidious deteriorations in patient status. Alerts are propagated via a multi-stakeholder workflow involving patients, family caregivers, general practitioners, non-profit organizations, and local health authorities, who interact with the platform through role-specific dashboards that support longitudinal visualization, risk stratification, and integration with existing electronic health record infrastructures. A pilot case study in a tele-home care programme for elderly patients in Calabria demonstrates the feasibility and potential of this architecture to improve safety, foster patient engagement, and strengthen continuity of care in geographically isolated communities, offering a scalable blueprint for territorial telemedicine in similar rural and mountainous contexts.
- 2 February 2026
- Published by MDPI AG in Preprints.org
Open AccessReal-Time Remote Patient Monitoring in Mountainous Regions: Architecture and Pilot Deployment of a Telemedicine Ecosystem
The rugged morphology and dispersed settlements of the Calabrian region pose long-standing barriers to timely and equitable access to healthcare, particularly for elderly and fragile populations living in mountainous areas. In this paper, we present a telemedicine ecosystem specifically designed for Calabria that integrates certified wearable wristbands, secure communication infrastructure, and intelligent back-end services to enable continuous home monitoring and rapid clinical intervention. Building on the SidlyCare platform, the system acquires real-time physiological signals (heart rate and oxygen saturation), streams them to a central server through encrypted channels, and applies machine learning and deep learning-based anomaly detection models to identify both acute and insidious deteriorations in patient status. Alerts are propagated via a multi-stakeholder workflow involving patients, family caregivers, general practitioners, non-profit organizations, and local health authorities, who interact with the platform through role-specific dashboards that support longitudinal visualization, risk stratification, and integration with existing electronic health record infrastructures. A pilot case study in a tele-home care programme for elderly patients in Calabria demonstrates the feasibility and potential of this architecture to improve safety, foster patient engagement, and strengthen continuity of care in geographically isolated communities, offering a scalable blueprint for territorial telemedicine in similar rural and mountainous contexts.