Mohammed Cherkaoui

Background/Objectives: Accurate drug-target affinity (DTA) prediction supports virtual screening, lead optimization, and drug repurposing. This study investigates whether replacing the conventional post-aggregation multilayer perceptron within a Graph Isomorphism Network drug encoder with a Kolmogorov-Arnold Network (KAN) transformation improves DTA prediction. Methods: The resulting KAN-GIN framework combines molecular-graph representations with a convolutional protein-sequence encoder. It was evaluated on the DAVIS, KIBA, and METZ benchmarks against a matched GIN baseline and representative multiscale and graph-pretraining-based DTA architectures under a common split and evaluation protocol. Complementary analyses examined latent-space organization, gradient-based molecular attribution, and an enhancer of zeste homolog 2 (EZH2) candidate-ranking case study. Results: KAN-GIN consistently reduced prediction error relative to the matched GIN baseline and achieved competitive performance against the additional architectures, including the strongest overall results on METZ. The latent-space findings were dataset-dependent, while gradient-based attribution identified molecular substructures contributing to individual predictions. The EZH2 case study illustrated how KAN-GIN can support candidate ranking before structure-based evaluation. Conclusions: KAN-based post-aggregation transformations represent an emerging design option for graph-based DTA prediction. However, the latent-space and structure-based findings remain computational and should not be interpreted as evidence of mechanistic biological relevance without prospective experimental validation.

Automated segmentation of the left and right ventricles (LVs and RVs) in cine cardiac MRI (CMR) underpins reliable volumetry and mass estimation. However, papillary muscles and trabeculae (PM/T) introduce clinically meaningful variability and exacerbate cross-dataset domain shift. We present VentrEX, an anatomically guided pipeline. The core segmenter, VentrEX-Seg, is a 3D encoder-decoder with parallel channel-spatial attention and a Transformer bottleneck. Training is performed exclusively on ACDC. A lightweight PM/T module automatically extracts papillary and trabecular burdens and standardizes cavity volumes. External evaluation is zero-shot (no fine-tuning) on Sunnybrook (LV) and MM-WHS MRI (RV). We report Dice, HD95 (mm); for volumetry, we use Bland-Altman analyses (LV and RV volumes). Attention/Grad-CAM visualizations support interpretability. On ACDC, VentrEX achieved higher Dice and lower boundary error than U-Net, nnU-Net, CBAM, and VentrEX-Seg. Zero-shot performance was preserved externally (e.g., Sunnybrook LV Dice 0.9053, HD95 4.95 mm; MM-WHS RV Dice 0.9236, HD95 6.61 mm). Patient-level Bland–Altman analyses characterized LV and RV volumetric agreement. Qualitative overlays and 3D reconstructions showed fewer PM/T “leaks” and anatomically plausible borders across ED/ES. Single-source training with dual zero-shot external tests demonstrates robustness under domain shift. The combination of parallel attention and a Transformer bottleneck enables accurate, transparent cine-CMR segmentation across datasets.

The main protease (MPro) of coronaviruses (CoVs) is an essential enzyme involved in viral replication and represents an attractive target for antiviral drug discovery. Based on the similar binding pocket residues within the MPro of different CoVs, this study aimed to identify potential inhibitors of SARS-CoV-2 MPro from PDB ID 6M2N using integrated computational approaches. Interaction-based pharmacophore modeling, virtual screening, molecular docking, MM-GBSA binding energy calculation, and molecular dynamics simulation (MDS) were performed using BIOVIA Discovery Studio. The validated pharmacophore model was utilized to screen the ZINC database, followed by docking and 100 ns MDS analyses of the top-ranked compounds. The pharmacophore model 01 demonstrated favorable predictive performance (AUC = 0.781). Virtual screening identified 483 compounds, from which 15 compounds were selected for docking studies. Among them, ZINC95473654 (Lig-1), ZINC95473725 (Lig-2), and ZINC08792368 (Lig-3) exhibited strong binding affinity toward MPro. Lig-1 demonstrated the best docking score and binding free energy, along with stable interactions with key catalytic residues HIS41, CYS145, and GLU166. MDS analyses further confirmed that Lig-1, Lig-2 and Lig-3 maintained stable conformations. The hydrogen bond distance monitoring and post MDS-MM-GBSA results suggest Lig-1 followed by Lig-3 as an inhibitor for MPro and persistent intermolecular interactions throughout the 100 ns simulation period. The findings suggest that Lig-1, followed by Lig-3, may serve as promising computational lead compounds targeting SARS-CoV-2 MPro, representing promising candidates for further experimental validation.

Parvovirus infection causes severe diseases in both feline and canine species. It primarily affects adult cats and dogs but poses a higher risk to kittens and puppies. This virus is highly contagious and is easily transmitted through contaminated food, shared shelter environments, as well as the hands and clothing of people. The recovered species may continue to shed parvovirus in their feces for an extended period, leading to severe environmental contamination. There is no universal vaccine available that protects dogs and cats against parvovirus infections. The main goal of this study is to design a pan-parvovirus multiepitope-based vaccine that could be administered to dogs and cats. We utilized AI-machine learning-incorporated server tools such as IEDB and NetMHCpan to predict B-cell and T-cell epitopes. VaxiJen and ToxinPred were used to analyze immune characteristic features and docking with feline alleles using the HADDOCK server. Following this, the immune response and stability of the vaccine construct were confirmed with disulfide engineering, normal mode analysis, and molecular docking performed with toll-like receptors of both feline and canine (TLR4 and TLR5), and molecular dynamics simulation was performed for 10 ns. The triggered immune response was determined with immuno-simulation (ImmSim), and their activity in a biological environment was reinforced with in silico cloning. The B-cell epitopes (NS1-9, NS2-4, VP1-12 and VP2-9) predicted with the IEDB database were subjected to antigenicity prediction. MHC class I and IFN prediction and MHC class II and IL-4 prediction were performed with IEDB and NetMHCpan. The T-cell epitopes showed high binding affinities with the feline alleles. The final vaccine was designed by combining the top-ranked B-cell epitopes and T-cell epitopes, filtered for high antigenicity, non-allergic, non-toxic, and good solubility, and with the better binding affinity score of the structural and non-structural proteins (NS1, NS2, VP1, and VP2) of feline and canine parvoviruses through linkers and adjuvants. The disulfide bond prediction and normal mode analysis showed that our vaccine construct is stable and flexible. The molecular docking analysis was performed between the designed vaccine epitopes and the TLRs (TLR4–feline and TLR5–canine) with Biovia Discovery Studio using Zdock; it showed better binding interactions with a value of 22.26 (Zdock score), −47.409 (Zrank score) for feline and 16.54 (Zdock score), −134.295 (Zrank score) for canine. A pan-multi-epitope-based vaccine based on the two structural and non-structural proteins (NS1, NS2, VP1, and VP2) was designed and constructed to provide dual protection against parvovirus in both feline and canine species. The molecular docking and molecular dynamics simulation analysis showed higher binding affinities and stable conformations with canine (TLR5) and feline (TLR4) toll-like receptors. Although computational analysis supports the prediction of top-ranked epitopes and their immunogenic properties with greater precision, further experimental validation is required before they can be used against these viruses.

Background/Objectives: Antibiotic resistance in bacteria poses a major health problem worldwide. Therefore, to counteract this life-threatening problem, we sought to investigate in the present study the possible potentiation of the efficacy of amoxicillin (AMX) and clavulanic acid (CA) by 1,8-cineole (CN), a candidate resistance-modulating agent. The approach seeks to investigate, in vitro and in silico, the interactions among these three molecules. Methods: The antibacterial activity was determined against resistant Escherichia coli (E. coli) using microdilution methods, synergy tests, and time-kill assays for AMX, CA, and CN, used either separately or in combinations: AMX–CA, AMX–CN, CA–CN, and AMX–CA–CN. Furthermore, an in silico drug design methodology was employed, utilizing an integrated workflow that combines Density Functional Theory (DFT) for ligand optimization with molecular docking simulations to evaluate binding energies and interactions between the penicillin-binding protein (PBP) and ligands. Results: In vitro synergy experiments and time-kill assays revealed substantial antibacterial efficacy of the AMX–CA–CN combination. In silico analyses, performed under the simplifying assumption of a pre-assembled multi-ligand entity, were consistent with these findings: within our docking model, the AMX–CA–CN combination exhibited the most favorable computed binding affinity to a representative penicillin-binding protein (PBP3, PDB 7ONW). Conclusions: 1,8-cineole can potentiate the antibacterial effects of AMX–CA, indicating that the AMX–CA–CN combination warrants further evaluation as a candidate adjunctive strategy against ESBL-producing E. coli.

Background. The main protease (MPro) of coronaviruses (CoVs) is an essential enzyme involved in viral replication and represents an attractive target for antiviral drug discovery. Based on the similar binding pocket residues within the MPro of different CoVs, the study aimed to identify potential inhibitors of SARS-CoV-2 MPro from PDB ID 6M2N, using integrated computational approaches. Methods. Interaction-based pharmacophore modeling, virtual screening, molecular docking, MM-GBSA binding energy calculation, and molecular dynamics (MD) simulation were performed using BIOVIA Discovery Studio. The validated pharmacophore model was utilized to screen the ZINC database, followed by docking and 100 ns MD simulation analyses of the top-ranked compounds. Results. The pharmacophore model 01 demonstrated favourable predictive performance (AUC = 0.781). Virtual screening identified 483 compounds, from which 21 compounds were selected for docking studies. Among them, ZINC95473654 (Lig-1), ZINC95473725 (Lig-2), and ZINC08792368 (Lig-3) exhibited strong binding affinity toward MPro. Lig-1 demonstrated the best docking score and binding free energy along with stable interactions with key catalytic residues HIS41, CYS145, and GLU166. MD simulation analyses further confirmed that Lig-1, Lig-2 and Lig-3 maintained stable conformations. The hydrogen bond distance monitoring and post MD-MM-GBSA results suggest Lig-1 followed by Lig-3 as an inhibitor for Mpro and persistent intermolecular interactions throughout the 100 ns simulation period. Conclusion. The findings suggest that Lig-1, followed by Lig-3, may serve as promising computational lead compounds targeting SARS-CoV-2 Mpro, representing promising candidates for further experimental validation.

Background/Objectives: Parvovirus infection cause sever diseases in both feline and canine species, mostly affects adult cats and dogs, but cause higher risk in the kittens and puppies. This virus is known to be contagious; the simple way of spreading occurs through food and shelter as well as hands and clothing of people. The recovered species may continue to shed parvovirus in their feces for an extended period, leading to severe environmental contamination. There is no universal vaccine available that protect dogs and cats against parvovirus infections. The main goal of this study is to design a pan parvovirus multiepitope DNA based vaccine that could be administered to dog and cats. Methods: We utilized AI-machine learning incorporated server tools such as IEDB and NetMHCpan to predict B-cell and T-cell epitopes. VaxiJen and ToxinPred were used to analyze immune characteristic features and docking with feline alleles using HADdock server. Following, the immune response and stability of vaccine construct was confirmed with disulfide engineering, Normal mode analysis and molecular docking, and dynamic simulations were done with Toll like receptors of both feline and canine (TLR4 and TLR5) for 50ns. The triggered immune response was determined with immuno-simulation (ImmSim) and their activity in biological environment was reinforced with in silico cloning. Results: The B cell epitopes (NS1 - 9, NS2 - 4, VP1- 12 and VP2- 9) predicted with IEDB database were subjected to antigenicity prediction. MHC class I and IFN prediction and MHC class II and IL-4 prediction were done with IEDB and NetMHCpan. The T cell epitopes showed high binding affinities with the feline alleles. The final vaccine was designed by combining the top-ranked B-cell epitopes T- cell epitopes, filtered from high antigenic, non-allergic, non-toxic and good solubility, and with the better binding affinity score of the structural and non- structural proteins (NS1, NS2, VP1 and VP2) of feline and canine parvoviruses through linkers and adjuvants. The disulfide bond prediction and Normal mode analysis showed our vaccine construct are stable and flexible. The molecular docking analysis was performed between the designed vaccine epitopes and the TLRs (TLR4 – feline and TLR5 – canine) with Biovia Discovery Studio using Zdocker, it showed the better binding interaction with value of 22.26 (Zdock score), -47.409 (Zrank score) for feline and 16.54 (Zdock score), -134.295 (Zrank score) for canine. Conclusions: The pan multi-epitope DNA based vaccine combining the four major structural proteins (NS1, NS2, VP1 and VP2) possess dual purpose to protect both feline and canine species against the parvovirus were designed and constructed. The molecular docking and dynamic simulation analysis showed higher binding affinities and stable conformation with canine (TLR5) and feline (TLR4) toll like receptors. Though computational analysis will support us to predict the more precise top-ranked epitopes and their immuno-antigenic properties, further experimental validation will be required to be used against those viruses.

Background. The main protease (Mpro/3CLpro) of coronaviruses (CoVs) is an essential enzyme involved in viral replication and represents an attractive target for antiviral drug discovery. Based on the similar binding pocket residues within the Mpro of different CoVs, the study aimed to identify potential inhibitors of Mpro from PDB ID 6M2N, using inte-grated computational approaches. Methods. Structure-based pharmacophore modeling, virtual screening, molecular docking, MM-GBSA binding energy calculation, and molecu-lar dynamics (MD) simulation were performed using BIOVIA Discovery Studio. The vali-dated pharmacophore model was utilized to screen the ZINC database, followed by dock-ing and 100 ns MD simulation analyses of the top-ranked compounds. Results. The pharmacophore model 01 demonstrated favourable predictive performance (AUC = 0.781). Virtual screening identified 483 compounds, from which 21 compounds were selected for docking studies. Among them, ZINC95473654 (Lig-1), ZINC95473725 (Lig-2), and ZINC08792368 (Lig-3) exhibited strong binding affinity toward Mpro. Lig-1 demonstrated the best docking score and binding free energy along with stable interactions with key cat-alytic residues HIS41, CYS145, and GLU166. MD simulation analyses further confirmed that Lig-1 and Lig-2 maintained stable conformations and persistent intermolecular inter-actions throughout the 100 ns simulation period. Conclusion. The findings suggest that Lig-1, followed by Lig-2, may serve as promising CoVs Mpro inhibitors and warrant fur-ther experimental validation. Further experimental validation are required to consolidate the identified compounds as universal inhibitors of the CoVs-Mpro enzyme.

Background/Objectives: Parvovirus infection cause sever diseases in both feline and ca-nine species, mostly affects adult cats and dogs, but cause higher risk in the kittens and puppies. This virus is known to be contagious; the simple way of spreading occurs through food and shelter as well as hands and clothing of people. The recovered species may continue to shed parvovirus in their feces for an extended period, leading to severe environmental contamination. There is no universal vaccine available that protect dogs and cats against parvovirus infections. The main goal of this study is to design a pan parvovirus multiepitope DNA based vaccine that could be administered to dog and cats. Methods: We utilized AI-machine learning incorporated server tools such as IEDB and NetMHCpan to predict B-cell and T-cell epitopes. VaxiJen and ToxinPred were used to analyze immune characteristic features and docking with feline alleles using HADdock server. Following, the immune response and stability of vaccine construct was confirmed with disulfide engineering, Normal mode analysis and molecular docking, and dynamic simulations were done with Toll like receptors of both feline and canine (TLR4 and TLR5) for 50ns. The triggered immune response was determined with immuno-simulation (ImmSim) and their activity in biological environment was reinforced with in silico cloning. Results: The B cell epitopes (NS1 - 9, NS2 - 4, VP1- 12 and VP2- 9) predicted with IEDB database were subjected to antigenicity prediction. MHC class I and IFN prediction and MHC class II and IL-4 prediction were done with IEDB and NetMHCpan. The T cell epitopes showed high binding affinities with the feline alleles. The final vaccine was de-signed by combining the top-ranked B-cell epitopes T- cell epitopes, filtered from high antigenic, non-allergic, non-toxic and good solubility, and with the better binding affinity score of the structural and non- structural proteins (NS1, NS2, VP1 and VP2) of feline and canine parvoviruses through linkers and adjuvants. The disulfide bond prediction and Normal mode analysis showed our vaccine construct are stable and flexible. The molecular docking analysis was performed between the designed vaccine epitopes and the TLRs (TLR4 – feline and TLR5 – canine) with Biovia Discovery Studio using Zdocker, it showed the better binding interaction with value of 22.26 (Zdock score), -47.409 (Zrank score) for feline and 16.54 (Zdock score), -134.295 (Zrank score) for canine. Conclusions: The pan multi-epitope DNA based vaccine combining the four major structural proteins (NS1, NS2, VP1 and VP2) possess dual purpose to protect both feline and canine species against the parvovirus were designed and constructed. The molecular docking and dynamic simulation analysis showed higher binding affinities and stable conformation with canine (TLR5) and feline (TLR4) toll like receptors. Though computational analysis will support us to predict the more precise top-ranked epitopes and their immuno-antigenic properties, further experimental validation will be required to be used against those viruses.

The limited selectivity of most catalytic-site ADAMTS-5 inhibitors and the necessity to preserve aggrecan integrity in early-grade knee osteoarthritis require the development of selective aggrecanase inhibitors. The present study conducted rational in silico screening of flavonoids as potential ADAMTS-5 inhibitors by integrating high-throughput virtual screening, molecular docking, and molecular dynamic simulations targeting the exosite domains of ADAMTS-5 (the Disintegrin-like and spacer domain). The objective was to identify plant-derived flavonoids with favorable drug-like properties and specific interactions towards the ADAMTS-5 exosite as a more targeted alternative to catalytic-site inhibition. In this study, 847 flavonoids were screened using drug-likeness and ADME criteria to identify promising leads. The top 16 selected flavonoids were further subjected to molecular docking and SAR analysis. Of these compounds, Homoeriodictyol satisfied key drug-likeness criteria and exhibited the highest binding affinity to the Disintegrin-like domain, with a binding energy of −23.1 kcal/mol and favorable interactions. Molecular dynamics simulations of the Homoeriodictyol–ADAMTS-5 complex over 100 ns demonstrated stable binding throughout the trajectory. DCCM analysis and PCA further supported the proposed exosite-mediated modulation. To extend exosite mapping beyond the Disintegrin-like domain, this study also examined the spacer domain using a machine-learning-predicted structural model and identified key residues that contribute to ligand binding.