Biology subjects

Lio', P.

Publications and source records attributed to Lio', P..

10 recordsLinked to original sources

bioRxiv2026

FibrilNet maps conserved and tissue-specific molecular environments across systemic amyloidoses

Systemic amyloidoses are initiated by distinct amyloidogenic precursor proteins but frequently contain recurrent extracellular, complement, lipid-transport and matrix-remodelling components. Whether these recurrent proteins form a conserved systems-level environment across amyloid diseases, and how strongly that environment depends on precursor and tissue context, remains unresolved. We developed FibrilNet, a network framework that integrates experimentally defined amyloid proteomes with a human protein protein interaction graph and Gene Ontology derived semantic information. FibrilNet compares topology-only random walk with restart (RWR) with ontology aware semantic RWR in frozen leave-one-out module reconstruction and precursor-seeded prioritization tasks. The human graph contains 17,997 proteins and 925,977 physical interactions, with a 9-dimensional semantic representation of interaction context. In expanded cardiac transthyretin amyloidosis (ATTR), semantic-RWR increased mean reciprocal rank (MRR) from 0.00167 to 0.05015 and Recall@100 from 0.0199 to 0.3377, improving 132 of 151 held-out targets. Significant semantic gains were also observed in renal serum amyloid A amyloidosis (AA) and leukocyte chemotactic factor 2 amyloidosis (ALECT2). Across compact ATTR, light-chain amyloidosis (AL), AA and ALECT2 modules, APCS, VTN and TIMP3 formed a direct four-disease recurrent core, while APOE occurred in three of four modules. A tissue-aware ATTR analysis showed limited overlap between cardiac and neurologic modules (19 shared proteins; Jaccard 0.0569). In the hTTR-A97S peripheral-nerve model, semantic-RWR significantly improved reconstruction of the 202-protein mapped neurologic module, with the strongest evidence concentrated in the downregulated proteomic program. TTR-seeded propagation improved with semantic information but remained weak in absolute terms, separating precursor identity from the distributed downstream molecular environment. These results support a multilayer model in which a restricted conserved amyloid environment coexists with precursor-, tissue- and disease-specific organization

bioinformatics↗

bioRxiv2026

The illusion of interpretability in biologically informed neural networks

Biologically informed neural networks (BINNs), also known as visible neural networks (VNNs), are widely adopted in omics because their architectures mirror known biological structures, such as gene-to-pathway relationships, and are therefore often assumed to be inherently interpretable. This assumption implies that learned gene-to-pathway weights and pathway node activations reflect meaningful biological mechanisms. Here, we show that this premise fails for a classical reason: nonidentifiability. Using a controlled teacher and student framework, we demonstrate that even under ideal conditions, including noiseless data, the correct model class, and identical sparse wiring, a BINN can perfectly recover the input-to-output mapping while failing to recover both gene-to-pathway weights and pathway activations. This failure persists across classification, regression, and survival tasks, and remains robust to variations in biological structure and network depth. Thus, the problem is not merely overparameterization or poor optimization: learning from outputs alone does not identify internal structure. Since biological mechanisms are not directly observed, recovering them from predictions alone is harder, not easier, than recovering neural network parameters, which are already known to be nonidentifiable. Critically, this failure reflects standard practice: widely used BINNs do not impose objective level constraints on gene-to-pathway weights or pathway activations, and therefore operate precisely in the regime modeled by our teacher-student framework. These results indicate that architectural transparency does not imply mechanistic interpretability. Without constraints that explicitly enforce identifiability, the apparent interpretability of BINNs reflects their design rather than what they actually learn.

bioinformatics↗

bioRxiv2025

Structure-based Generation of a Secondary Nucleation Inhibitor in alpha-Synuclein Aggregation Using a Conditional Diffusion Model

The process of -synuclein aggregation results in the formation of amyloid fibrils, which accumulate in the brain of patients affected by Parkinsons disease. Among possible therapeutic strategies to cure this disease, one approach is based on the development of compounds capable of inhibiting -synuclein aggregation. An effective inhibition could be achieved by blocking the nucleation sites on the surface of the amyloid fibrils that are responsible for their autocatalytic proliferation. Here, we report a strategy based on deep learning to achieve this goal, which uses an E(3)-equivariant conditional diffusion model. By using this approach, we designed and tested experimentally candidate small molecules. We found that one of these small molecules acts as a potent inhibitor of secondary nucleation in -synuclein aggregation. These results provide evidence that generative diffusion models offer effective tools for drug design.

biophysics↗

bioRxiv2024

UnifiedGreatMod: A New Holistic Modeling Paradigm for Studying Biological Systems on a Complete and Harmonious Scale

MotivationComputational models are crucial for addressing critical questions about systems evolution and deciphering system connections. The pivotal feature of making this concept recognisable from the biological and clinical community is the possibility of quickly inspecting the whole system, bearing in mind the different granularity levels of its components. This holistic view of system behaviour expands the evolution study by identifying the heterogeneous behaviours applicable, for example, to the cancer evolution study. ResultsTo address this aspect, we propose a new modelling paradigm, UnifiedGreatMod, which allows modellers to integrate fine-grained and coarse-grained biological information into a unique model. It allows for functional studies, in which the understanding of the systems multi-level stable condition and the systems fluctuating condition are combined to investigate the functional dependencies among the biological entities in the system under study. This is achieved thanks to the hybridisation of two analysis approaches that capture a systems different granularity levels. The proposed paradigm was then implemented into the open-source, general modelling framework GreatMod, in which a graphical meta-formalism is exploited to simplify the model creation phase and R languages to define user-defined analysis workflows. The proposals effectiveness was demonstrated by mechanistically simulating the metabolic output of Echerichia coli under environmental nutrient perturbations and integrating a gene expression dataset. Additionally, the UnifiedGreatMod was used to examine the responses of luminal epithelial cells to Clostridium difficile infection.

systems biology↗

bioRxiv2021

Discovering cancer driver genes and pathways using stochastic block model graph neural networks

The identification of genes and pathways responsible for the transformation of normal cells into malignant ones represents a pivotal step to understand the aetiology of cancer, to characterise progression and relapse, and to ultimately design targeted therapies. The advent of high-throughput omic technologies has enabled the discovery of a significant number of cancer driver genes, but recent genomic studies have shown these to be only necessary but not sufficient to trigger tumorigenesis. Since most biological processes are the results of the interaction of multiple genes, it is then conceivable that tumorigenesis is likely the result of the action of networks of cancer driver and non-driver genes. Here we take advantage of recent advances in graph neural networks, combined with well established statistical models of network structure, to build a new model, called Stochastic Block Model Graph Neural Network (SBM-GNN), which predicts cancer driver genes and cancer mediating pathways directly from high-throughput omic experiments. Experimental analysis of synthetic datasets showed that our model can correctly predict genes associated with cancer and recover relevant pathways, while outperforming other state-of-the-art methods. Finally, we used SBM-GNN to perform a pan-cancer analysis, where we found genes and pathways directly involved in the hallmarks of cancer controlling genome stability, apoptosis, immune response, and metabolism.

bioinformatics↗

bioRxiv2021

Multimodal image fusion via deep generative models

Recently, it has become progressively more evident that classic diagnostic labels are unable to accurately and reliably describe the complexity and variability of several clinical phenotypes. This is particularly true for a broad range of neuropsychiatric illnesses such as depression and anxiety disorders or behavioural phenotypes such as aggression and antisocial personality. Patient heterogeneity can be better described and conceptualized by grouping individuals into novel categories, which are based on empirically-derived sections of intersecting continua that span both across and beyond traditional categorical borders. In this context, neuroimaging data carry a wealth of spatiotemporally resolved information about each patients brain. However, they are usually heavily collapsed a priori through procedures which are not learned as part of model training, and consequently not optimized for the downstream prediction task. This is due to the fact that every individual participant usually comes with multiple whole-brain 3D imaging modalities often accompanied by a deep genotypic and phenotypic characterization, hence posing formidable computational challenges. In this paper we design and validate a deep learning architecture based on generative models rooted in a modular approach and separable convolutional blocks (which result in a 20-fold decrease in parameter utilization) in order to a) fuse multiple 3D neuroimaging modalities on a voxel-wise level, b) efficiently convert them into informative latent embeddings through heavy dimensionality reduction, c) maintain excellent generalizability and minimal information loss. As proof of concept, we test our architecture on the well characterized Human Connectome Project database (n=974 healthy subjects), demonstrating that our latent embeddings can be clustered into easily separable subject strata which, in turn, map to extremely different phenotypical information (including organic, neuropsychological, personality variables) which was not included in the embedding creation process. The ability to extract meaningful and separable phenotypic information from brain images alone can aid in creating multi-dimensional biomarkers able to chart spatio-temporal trajectories which may correspond to different pathophysiological mechanisms unidentifiable to traditional data analysis approaches. In turn, this may be of aid in predicting disease evolution as well as drug response, hence supporting mechanistic disease understanding and also empowering clinical trials.

neuroscience↗

bioRxiv2019

Adversarial generation of gene expression data

The problem of reverse engineering gene regulatory networks from high-throughput expression data is one of the biggest challenges in bioinformatics. In order to benchmark network inference algorithms, simulators of well-characterized expression datasets are often required. However, existing simulators have been criticized because they fail to emulate key properties of gene expression data. In this study we address two problems. First, we propose mechanisms to faithfully assess the realism of a synthetic gene expression dataset. Second, we design an adversarial simulator of expression data, gGAN, based on a Generative Adversarial Network. We show that our model outperforms existing simulators by a large margin, achieving realism scores that are up to 17 times higher than those of GeneNetWeaver and SynTReN. More importantly, our results show that gGAN is, to our best knowledge, the first simulator that passes the Turing test for gene expression data proposed by Maier et al. (2013).

bioinformatics↗

bioRxiv2019

Variational Autoencoders for Cancer Data Integration: Design Principles and Computational Practice

International initiatives such as the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) are collecting multiple data sets at different genome-scales with the aim to identify novel cancer bio-markers and predict patient survival. To analyse such data, several machine learning, bioinformatics and statistical methods have been applied, among them neural networks such as autoencoders. Although these models provide a good statistical learning framework to analyse multi-omic and/or clinical data, there is a distinct lack of work on how to integrate diverse patient data and identify the optimal design best suited to the available data.\n\nIn this paper, we investigate several autoencoder architectures that integrate a variety of cancer patient data types (e.g., multi-omics and clinical data). We perform extensive analyses of these approaches and provide a clear methodological and computational framework for designing systems that enable clinicians to investigate cancer traits and translate the results into clinical applications. We demonstrate how these networks can be designed, built and, in particular, applied to tasks of integrative analyses of heterogeneous breast cancer data. The results show that these approaches yield relevant data representations that, in turn, lead to accurate and stable diagnosis.

bioinformatics↗

bioRxiv2019

PECLIDES Neuro - A Personalisable Clinical Decision Support System for Neurological Diseases

Neurodegenerative diseases such as Alzheimers and Parkinsons impact millions of people worldwide. Early diagnosis has proven to greatly increase the chances of slowing down the diseases progression. Correct diagnosis often relies on the analysis of large amounts of patient data, and thus lends itself well to support from machine learning algorithms, which are able to learn from past diagnosis and see clearly through the complex interactions of a patients symptoms. Unfortunately, many contemporary machine learning techniques fail to reveal details about how they reach their conclusions, a property considered fundamental when providing a diagnosis. This is one reason why we introduce our Personalisable Clinical Decision Support System PECLIDES that provides a clear insight into the decision making process on top of the diagnosis. Our algorithm enriches the fundamental work of Masheyekhi and Gras in data integration, personal medicine, usability, visualisation and interactivity.\n\nOur decision support system is an operation of translational medicine. It is based on random forests, is personalisable and allows a clear insight into the decision making process. A well-structured rule set is created and every rule of the decision making process can be observed by the user (physician). Furthermore, the user has an impact on the creation of the final rule set and the algorithm allows the comparison of different diseases as well as regional differences in the same disease1.

bioinformatics↗

bioRxiv2018

Early Detection of Neurological Dysfunction Using Blood Cell Transcript Profiles

Identification of genes whose regulation of expression is similar in both brain and blood cells could enable monitoring of significant neurological traits and disorders by analysis of blood samples. We thus employed transcriptional analysis of pathologically affected tissues, using agnostic approaches to identify overlapping gene functions and integrating this transcriptomic information with expression quantitative trait loci (eQTL) data. Here, we estimate the correlation of genetic expression in the top-associated cis-eQTLs of brain tissue and blood cells in Parkinsons (PD). We introduced quantitative frameworks to reveal the complex relationship of various biasing genetic factors in PD, a neurodegenerative disease. We examined gene expression microarray and RNA-Seq datasets from human brain and blood tissues from PD-affected and control individuals. Differentially expressed genes (DEG) were identified for both brain and blood cells to determine common DEG overlaps. Based on neighborhood-based benchmarking and multilayer network topology aproaches we then developed genetic associations of factors with PD. Overlapping DEG sets underwent gene enrichment using pathway analysis and gene ontology methods, which identified candidate common genes and pathways. We identified 12 significantly dysregulated genes shared by brain and blood cells, which were validated using dbGaP (gene SNP-disease linkage) database for gold-standard benchmarking of their significance in disease processes. Ontological and pathway analyses identified significant gene ontology and molecular pathways that indicate PD progression. In sum, we found possible novel links between pathological processes in brain and blood cells by examining cell path-way commonalities, corroborating these associations using well validated datasets. This demonstrates that for brain-related pathologies combining gene expression analysis and blood cell cis-eQTL is a potentially powerful analytical approach. Thus, our methodologies facilitate data-driven approaches that can advance knowledge of disease mechanisms and may enable prediction of neurological dysfunction using blood cell transcript profiling.

bioinformatics↗