Biology subjects

Gentile, F.

Publications and source records attributed to Gentile, F..

11 recordsLinked to original sources

bioRxiv2026

Interconnecting ADC Structure with Tumor Cell Biology with Multimodal Learning

Antibody-drug conjugates (ADCs) represent a significant advancement in cancer therapy, yet their development remains constrained by high attrition rates driven by an incomplete understanding of how ADC chemical design interconnects with tumor biology. Compounding this challenge, the field has converged on a narrow set of redundant structural components, and current linker-payload systems that do not share a single mechanism of action. Existing drug response prediction frameworks cannot resolve this multidimensional complexity, relying predominantly on genomic inputs while protein-level biology is challenging to integrate. To address this, we developed a multimodal machine learning platform interconnecting ADC structural parameters with tumor cell biology across thousands of curated structure-activity datapoints, including multi-omics profiles from 1,479 human tumor cell lines and protein-level inputs from a unique model (GENCEP) that derives complete proteomic signatures. The model was validated through blinded retrospective evaluation and, critically, large coverage prospective prediction of cytotoxicity across 159 ADC-cell line combinations spanning five antigens, four mechanistically distinct linker-payload systems, and eight tumor types, most with no prior published associated ADC data. Performance surpassed industry benchmarks established for small molecule therapeutic modalities, demonstrating that protein-informed multimodal integrated framework is effective at capturing cytotoxic determinants at scale.

pharmacology and toxicology↗

bioRxiv2025

Harry Potter meets Markov: Neural event representation in the reading network during narrative processing

People segment ongoing information into meaningful mental representations known as events. Recent studies have found a spatio-temporal hierarchy in the event structure during movie watching and story listening. How the human brain segments stories while reading remains unclear. We examined event segmentation while reading Harry Potter using publicly available 3T fMRI data. To identify neural events, we employed a Hidden Markov Model (HMM) within six regions of interest (ROI) in the reading network. The ROIs included the inferior frontal gyrus (IFG), middle frontal gyrus (MFG), angular gyrus (AG), inferior frontal gyrus orbital (IFGorb), anterior temporal lobe (ATL), and posterior temporal lobe (PTL). These regions are also associated with DMN. The results revealed multiple timescales of events, with the IFG and AG representing shorter events, while the IFGorb, ATL, PTL and MFG represent relatively longer events. Boundaries of longer events were partially aligned within short events, suggesting a nested hierarchical event structure. Further comparisons identified associations between neural events and annotations from humans and ChatGPT (automated and less subjective segmentation) in the inferior frontal, temporal and parietal areas. Notably, the AG (a core region of the DMN and memory processing) and the IFG (relevant for language processing) exhibited stronger alignment with both sets of annotations. The findings suggest an important role of AG and IFG in event segmentation in reading. They also confirm the connection between the reading network and the DMN in event segmentation. This connection seems to be linked to the hierarchical processing of context-dependent semantic information during reading.

neuroscience↗

bioRxiv2025

Structure-based discovery of inhibitors of Mac1 domain of nonstructural protein-3 of SARS-CoV-2 by machine learning-augmented screening of chemical space

Significant efforts have been recently dedicated to the discovery of small molecule inhibitors against the Macrodomain 1 (Mac1) of nonstructural protein 3 (NSP3) as potential antivirals for SARS-CoV-2. Thus, Mac1 has also been selected as the target for the Critical Assessment of Hit-finding Experiments (CACHE) challenge #3. As contestants in that challenge, we developed a computational strategy that ranked on the top among all 23 participants in the competition and resulted in the discovery of a novel chemical series of non-charged Mac1 inhibitors. Those have been identified through the combination of machine learning-accelerated virtual screening of Enamine REAL Diversity Subset of approximately 25 million compounds and consequent hit expansion into the entire Enamine REAL Space library. In particular, the initially identified hit compound CACHE3-HI_1706_56 (KD = 20 {micro}M) was explored by probing 17 close analogues from a library of 44 billion molecules from the Enamine REAL. All those analogues effectively displaced the Mac1-binding ADP-ribose peptide, and 12 were confirmed to engage with Mac1 by the Surface Plasmon Resonance experiments, revealing a new chemical series of compounds for hit-to-lead optimization. The structure of the CACHE3-HI_1706_56-Mac1 complex was further determined at high resolution with crystallography, confirming initial computational predictions. Our results illustrate the effectiveness of ML-accelerated docking to rapidly identify novel chemical series and provide a strong foundation for the development of SARS-CoV-2 NSP3 Mac1 inhibitors.

biophysics↗

bioRxiv2025

A genetic signature predicts aggressive paraganglioma sensitivity to dual PI3K-CDK4/6 inhibition therapy

Effective medical therapies for metastatic paraganglioma (mPPGL) are currently lacking, leading to dismal prognosis. Building on our knowledge of molecular mechanisms driving PPGL progression, we assessed the therapeutic potential of targeting two critical processes: PI3K signaling and cell cycle regulation. The efficacy of buparlisib (PI3Ki) and ribociclib (CDK4/6i), individually and combined, was assessed in vitro using PPGL cell lines, rat- and patient-derived primary cells, and in vivo using PPGL cells-derived mouse xenografts. The combination therapy demonstrated superior antitumor activity compared to single agents, particularly in vivo. Mechanistically, the efficacy of the combination therapy was associated to the downregulation of FOXM1-controlled genes implicated in mitotic spindle assembly and chromosomal segregation, leading to mitotic catastrophe. Data mining and qRT-PCR showed this genetic signature to be upregulated in human mPPGLs. This suggests that aggressive PPGLs exhibit heightened vulnerability to dual PI3K and CDK4/6 inhibition, offering a promising therapeutic avenue for these challenging cancers.

cancer biology↗

bioRxiv2024

CACHE Challenge #1: targeting the WDR domain of LRRK2, a Parkinson's Disease associated protein.

The CACHE challenges are a series of prospective benchmarking exercises meant to evaluate progress in the field of computational hit-finding. Here we report the results of the inaugural CACHE #1 challenge in which 23 computational teams each selected up to 100 commercially available compounds that they predicted would bind to the WDR domain of the Parkinsons disease target LRRK2, a domain with no known ligand and only an apo structure in the PDB. The lack of known binding data and presumably low druggability of the target is a challenge to computational hit finding methods. Seventy-three of the 1955 procured molecules bound LRRK2 in an SPR assay with KD lower than 150 M and were advanced to a hit expansion phase where computational teams each selected up to 50 analogs each. Binding was observed in two orthogonal assays with affinities ranging from 18 to 140 M for seven chemically diverse series. The seven successful computational workflows varied in their screening strategies and techniques. Three used molecular dynamics to produce a conformational ensemble of the targeted site, three included a fragment docking step, three implemented a generative design strategy and five used one or more deep learning steps. CACHE #1 reflects a highly exploratory phase in computational drug design where participants sometimes adopted strikingly diverging screening strategies. Machine-learning accelerated methods achieved similar results to brute force (e.g. exhaustive) docking. First-in-class, experimentally confirmed compounds were rare and weakly potent, indicating that recent advances are not sufficient to effectively address challenging targets.

biochemistry↗

bioRxiv2024

Artificial intelligence as a ploy to delve into the intricate relationship between genetics and mitochondria in MASLD patients

Background and AimsMitochondrial (mt-) dysfunction is a hallmark of progressive MASLD. MtDNA copy number (mtDNA-CN) and cell-free circulating mtDNA (ccf-mtDNA), which reflect mt-mass and mt-dysfunction, respectively, are gaining attention as non-invasive disease biomarkers. We previously demonstrated that PNPLA3/MBOAT7/TM6SF2 deficiency in HepG2 cells increased mt-mass, mtDNA-CN and ccf-mtDNA. This study furtherly explored mt-biogenesis, function and mt-biomarkers in biopsied MASLD patients from a Discovery (n=28) and a Validation (n=824) cohort, stratified by the number of risk variants (NRV=3). We took advantage of artificial intelligence (AI) to develop new risk scores, predicting MASLD evolution by integrating anthropometric and genetic data (Age, BMI, NRV) with mt-biomarkers. MethodsHepatic mt-morphology and dynamics were assessed by TEM, IHC and gene expression. mtDNA-CN and ccf-mtDNA were measured in PBMCs and serum samples. GPT-4 was employed as AI tool to support the construction of novel risk scores for MASLD progressive forms (MASH, fibrosis and HCC). ResultsIn the Discovery cohort, NRV=3 patients showed the highest mt-mass and significant mt-morphological changes (i.e. membranes rupture). An elevated PGC-1, OPA1, DRP1 and PINK1, markers of mt-biogenesis, fusion and fission were found in these patients, supporting an enhanced mt-dynamics. However, PRKN protein levels were reduced, suggesting a premature block of mitophagy. In the Validation cohort, PGC-1 mRNA levels and mtDNA-CN were significantly higher in NRV=3 compared to patients with 1,2 or no variants. Circulating mtDNA-CN and ccf-mtDNA were augmented in NRV=3 patients and correlated with genetics and MASLD severity at multivariate analysis, supporting that both may independently modulate mt-dynamics and activity. By exploiting rsGPT-4 we then optimized the combination of non-invasive variables to get prediction models named Mitochondrial, Anthropometric, and Genetic Integration with Computational intelligence ("MAGIC-") for assessing MASH, fibrosis, and HCC, respectively. The MAGIC-MASH and MAGIC-Fib models showed AUCs of 73% and 76% in detecting MASH and fibrosis >1. Of note, MAGIC-HCC achieved an AUC of 86% (95% CI: 0.823-0.885), with 78.6% sensitivity and 81.5% specificity thus resulting the best score for the desired outcome. ConclusionsmtDNA-CN and ccf-mtDNA may have pathological and prognostic significance in MASLD patients, especially in those genetically-predisposed.

genetics↗

bioRxiv2024

Biochemical properties of chromatin domains define genome compartmentalization

Chromatin three-dimensional (3D) organization inside the cell nucleus determines the separation of euchromatin and heterochromatin domains. Their segregation results in the definition of active and inactive chromatin compartments, whereby the local concentration of associated proteins, RNA and DNA results in the formation of distinct subnuclear structures. Thus, chromatin domains spatially confined in a specific 3D nuclear compartment are expected to share similar epigenetic features and biochemical properties, in terms of accessibility and solubility. Based on this rationale, we developed the 4f-SAMMY-seq to map euchromatin and heterochromatin based on their accessibility and solubility, starting from as little as 10,000 cells. Adopting a tailored bioinformatic data analysis approach we reconstruct also their 3D segregation in active and inactive chromatin compartments and sub-compartments, thus recapitulating the characteristic properties of distinct chromatin states. A key novelty is the capability to map both the linear segmentation of open and closed chromatin domains, as well as their 3D compartmentalization in one single experiment.

genomics↗

bioRxiv2023

PROTACable is an Integrative Computational Pipeline of 3-D Modeling and Deep Learning to Automate the De Novo Design of PROTACs

Proteolysis-targeting chimeras (PROTACs) that engages two biological targets at once is a promising technology in degrading clinically relevant protein targets. Since factors that influence the biological activities of PROTACs are more complex than those of a small molecule drug, we explored a combination of computational chemistry and deep learning strategies to forecast PROTAC activity and enable automated design. A new method named PROTACable was developed for de novo design of PROTACs, which includes a robust 3-D modeling workflow to model PROTAC ternary complexes using a library of E3 ligase and linker and an SE(3)-equivariant graph transformer network to predict the activity of newly designed PROTACs. PROTACable is available at https://github.com/giaguaro/PROTACable/.

bioinformatics↗

bioRxiv2022

Deep brain cancer metastasis detection with wide-volume Raman spectroscopy through a single tapered fiber

We propose a low-invasive method to enable implantable, large volume Raman spectroscopy in arbitrarily deep brain regions of the mouse brain. Using a single tapered fiber as thin as 1 m at the tip, we identified diagnostic markers of brain metastasis - the most frequent brain tumor in human adults - with overall accuracy [≥] 90%. We view our approach as a promising complement to the existing palette of tools for optical interrogation of the brain.

neuroscience↗

bioRxiv2019

Deep Docking - a Deep Learning Approach for Virtual Screening of Big Chemical Datasets

Drug discovery is an extensive and rigorous process that requires up to 2 billion dollars of investments and more than ten years of research and development to bring a molecule "from bench to a bedside". While virtual screening can significantly enhance drug discovery workflow, it ultimately lags the current rate of expansion of chemical databases that already incorporate billions of purchasable compounds. This surge of available small molecules presents great opportunities for drug discovery but also demands for faster virtual screening methods and protocols. In order to address this challenge, we herein introduce Deep Docking (D2) - a novel deep learning-based approach which is suited for docking billions of molecular structures. The developed D2-platform utilizes quantitative structure-activity relationship (QSAR) based deep models trained on docking scores of subsets of a large chemical library (Big Base) to approximate the docking outcome for yet unprocessed molecular entries and to remove unfavorable structures in an iterative manner. We applied D2 to virtually screen 1.36 billion molecules form the ZINC15 library against 12 prominent target proteins, and demonstrated up to 100-fold chemical data reduction and 6,000-fold enrichment for top hits, without notable loss of well-docked entities. The developed D2 protocol can readily be used in conjunction with any docking program and was made publicly available.

bioinformatics↗

bioRxiv2017

Fast Periodic Stimulation (FPS): A Highly Effective Approach In fMRI Brain Mapping

Functional magnetic resonance imaging (fMRI) is a major technique for human brain mapping. We present a Fast Periodic Stimulation (FPS) fMRI approach, demonstrating its high effectiveness in defining category-selective brain regions. Observers see a dynamic stream of widely variable natural object images alternating at a fast rate (6 images/sec). Every 9 seconds, a short burst of variable face images contrasting with objects in pairs induces an objective 0.111 Hz face-selective neural response in the ventral occipito-temporal cortex and beyond. A model-free Fourier analysis achieves a two-fold increase in signal-to-noise ratio compared to a conventional block-design approach with identical stimuli. Periodicity of category contrast and random variability among images minimize low-level visual confounds while preserving naturalness of the stimuli, leading to the highest values (80-90%) of test-retest reliability yet reported in this area of research. FPS-fMRI opens a new avenue for understanding brain function with low temporal resolution methods.\n\nHighlightsFPS-fMRI achieves a two-fold increase in peak SNR over conventional approach\n\nFPS-fMRI reveals comprehensive extended face-selective areas including ATL\n\nFPS-fMRI achieves high specificity by minimizing influence of low-level visual cues\n\nFPS-fMRI achieves very high test-retest reliability (80%-90%) in spatial activation map\n\neTOC BlurbIn Brief\n\nGao et al. present a novel FPS-fMRI approach, which achieves a two-fold increase in peak signal-to-noise ratio in defining the neural basis of visual categorization while preserving ecological validity, minimizing low-level visual confounds and reaching very high (80%-90%) test-retest reliability.

neuroscience↗