Biology subjects

Tejera, E.

Publications and source records attributed to Tejera, E..

7 recordsLinked to original sources

bioRxiv2025

Viral Sentry AI (VirSentAI) - Automated Zoonotic Surveillance & Drug Repurposing Agent

Zoonotic viruses capable of jumping from animal reservoirs into human populations represent a persistent and unpredictable menace to global health. To confront this challenge, we developed Viral Sentry AI (VirSentAI), an autonomous agent designed to close the gap between viral emergence and therapeutic response. Unlike static analysis tools, VirSentAI operates as a continuous sentinel, automatically scanning public databases (e.g., NCBI) for new viral genomes and executing a three-stage agentic surveillance workflow, with distinct, specialized AI architectures for generated text, macromolecule sequences, and drug chemical data. First, the system is using gemma-2-9b, utilizing this Large Language Model to parse unstructured submission records and extract critical meta-information that provides context to the raw data. In the second stage, the system employs a novel deep-learning topology, virsentai-v2-hyena-dna-16k, a fine-tuned HyenaDNA model capable of processing complete viral genomes up to 160,000 bases. This architecture captures subtle, long-range genomic dependencies to predict human infectivity with high precision. Upon flagging a high-risk pathogen, the agent autonomously triggers a downstream therapeutic module as the stage three. It extracts NCBI viral protein sequences and utilizes a PLAPT (Protein-Ligand Affinity Prediction Transformer) model to calculate affinity interactions against the ChEMBL-curated set of approved therapeutics, instantly identifying candidates for drug repurposing. In rigorous cross-validation on a curated dataset of 31,728 complete viral genomes, the surveillance module demonstrated robust discriminatory power, achieving an AUROC of 0.95 in classifying human host potential. By integrating state-of-the-art genomic modeling with automated lead compound screening, VirSentAI offers a proactive, end-to-end solution for pandemic preparedness. The platform is freely accessible at https://muntisa.github.io/virsentai, with source code available at https://github.com/muntisa/virsentai.

bioinformatics↗

bioRxiv2024

NeoPhAR: Next-generation Phage Therapy for Antimicrobial Resistance

Growing antimicrobial resistance (AMR) infections are among the top contemporary concerns in public healthcare systems which can place considerable burdens on healthcare systems. Phage therapy has long been considered a viable option to help combat AMR infections, including on broad geographic scales. One bottleneck in the application of phage therapy is the accurate matching of host-specific phages to bacterial strains, which is traditionally done using molecular techniques. Here we present an open-access deep learning-based model that shows incredible potential to accurately predict phages for therapeutic use. Our goal is to attenuate the matching process so that specific phages, or a cocktail of phages can be prepared for patients with a high probability of therapeutic success, helping to democratize phage therapy treatments even in low to middle-income countries where genomic resources can be limited, costly, or time prohibitive. We feel this is an important first step towards incorporating and applying bioinformatic practices in promising fields such as phage therapy.

bioinformatics↗

bioRxiv2019

Prediction of breast cancer proteins using molecular descriptors and artificial neural networks: a focus on cancer immunotherapy proteins, metastasis driver proteins, and RNA-binding proteins

BackgroundBreast cancer (BC) is a heterogeneous disease characterized by an intricate interplay between different biological aspects such as ethnicity, genomic alterations, gene expression deregulation, hormone disruption, signaling pathway alterations and environmental determinants. Due to the complexity of BC, the prediction of proteins involved in this disease is a trending topic in drug design. MethodsThis work is proposing accurate prediction classifier for BC proteins using six sets of protein sequence descriptors and 13 machine learning methods. After using a univariate feature selection for the mix of five descriptor families, the best classifier was obtained using multilayer perceptron method (artificial neural network) and 300 features. ResultsThe performance of the model is demonstrated by the area under the receiver operating characteristics (AUROC) of 0.980 {+/-} 0.0037 and accuracy of 0.936 {+/-} 0.0056 (3-fold cross-validation). Regarding the prediction of 4504 cancer-associated proteins using this model, the best ranked cancer immunotherapy proteins related to BC were RPS27, SUPT4H1, CLPSL2, POLR2K, RPL38, AKT3, CDK3, RPS20, RASL11A and UBTD1; the best ranked metastasis driver proteins related to BC were S100A9, DDA1, TXN, PRNP, RPS27, S100A14, S100A7, MAPK1, AGR3 and NDUFA13; and the best ranked RNA-binding proteins related to BC were S100A9, TXN, RPS27L, RPS27, RPS27A, RPL38, MRPL54, PPAN, RPS20 and CSRP1. ConclusionsThis powerful model predicts several BC-related proteins which should be deeply studied to find new biomarkers and better therapeutic targets. The script and the results are available as a free repository at https://github.com/muntisa/neural-networks-for-breast-cancer-proteins.

bioinformatics↗

bioRxiv2019

Prediction of druggable proteins using machine learning and functional enrichment analysis: a focus on cancer-related proteins and RNA-binding proteins

BackgroundDruggable proteins are a trending topic in drug design. The druggable proteome can be defined as the percentage of proteins that have the capacity to bind an antibody or small molecule with adequate chemical properties and affinity. The screening and in silico modeling are critical activities for the reduction of experimental costs.\n\nMethodsThe current work proposes a unique prediction model for druggable proteins using amino acid composition descriptors of protein sequences and 13 machine learning linear and non-linear classifiers. After feature selection, the best classifier was obtained using the support vector machine method and 200 tri-amino acid composition descriptors.\n\nResultsThe high performance of the model is determined by an area under the receiver operating characteristics (AUROC) of 0.975 {+/-} 0.003 and accuracy of 0.929 {+/-} 0.006 (3-fold cross-validation). Regarding the prediction of cancer-associated proteins using this model, the best ranked druggable predicted proteins in the breast cancer protein set were CDK4, AP1S1, POLE, HMMR, RPL5, PALB2, TIMP1, RPL22, NFKB1 and TOP2A; in the cancer-driving protein set were TLL2, FAM47C, SAGE1, HTR1E, MACC1, ZFR2, VMA21, DUSP9, CTNNA3 and GABRG1; and in the RNA-binding protein set were PLA2G1B, CPEB2, NOL6, LRRC47, CTTN, CORO1A, SCAF11, KCTD12, DDX43 and TMPO.\n\nConclusionsThis powerful model predicts several druggable proteins which should be deeply studied to find better therapeutic targets and thus improve clinical trials. The scripts are freely available at https://github.com/muntisa/machine-learning-for-druggable-proteins.

bioinformatics↗

bioRxiv2019

ShadowCaster: compositional methods under the shadow of phylogenetic models for the detection of horizontal gene transfer events in prokaryotes.

Horizontal gene transfer (HGT) plays an important role in the evolution of many organisms, especially in prokaryotes where commonly occurs. Microbial communities can improve survival due to the evolutionary innovations induced by HGT events. Thus, several computational approaches have arisen to identify such events in recipient genomes. However, this has been proven to be a complex task due to the generation of a great number of false positives and the prediction disagreement among the existing methods. Phylogenetic reconstruction methods turned out to be the most reliable but they are not extensible to all genes/species and are computationally demanding when dealing with large datasets. On the other hand, the so-called surrogate methods that use heuristic solutions either based on nucleotide composition patterns or phyletic distribution of BLAST hits can be applied easily to genomic scale, however, they fail in identifying common HGT events. Here, we present ShadowCaster, a hybrid approach that sequentially combines compositional features under the shadow of phylogenetic models independent of tree reconstruction to improve the detection of HTG events in prokaryotes. ShadowCaster predicted successfully close and distant HTG events in both artificial and bacterial genomes. It detected HGT related to heavy metal resistance in the genome of Rhodanobacter denitrificans with higher accuracy than the most popular state-of-the-art computational approaches. ShadowCasters predictions showed the highest agreement among those obtained with other assayed methods. ShadowCaster is released as an open-source software under the GPLv3 license. Source code is hosted at https://github.com/dani2s/ShadowCaster and documentation at https://shadowcaster.readthedocs.io/en/latest/.

bioinformatics↗

bioRxiv2019

OncoOmics approaches to reveal essential genes in breast cancer: a panoramic view from pathogenesis to precision medicine

Breast cancer (BC) is a heterogeneous disease where each OncoOmics approach needs to be fully understood as a part of a complex network. Therefore, the main objective of this study was to analyze genetic alterations, signaling pathways, protein-protein interaction networks, protein expression, dependency maps and enrichment maps in 230 previously prioritized genes by the Consensus Strategy, the Pan-Cancer Atlas, the Pharmacogenomics Knowledgebase and the Cancer Genome Interpreter, in order to reveal essential genes to accelerate the development of precision medicine in BC. The OncoOmics essential genes were rationally filtered to 144, 48 (33%) of which were hallmarks of cancer and 20 (14%) were significant in at least three OncoOmics approaches: RAC1, AKT1 CCND1, PIK3CA, ERBB2, CDH1, MAPK14, TP53, MAPK1, SRC, RAC3, PLCG1, GRB2, MED1, TOP2A, GATA3, BCL2, CTNNB1, EGFR and CDK2. According to the Open Targets Platform, there are 111 drugs that are currently being analyzed in 3151 clinical trials in 39 genes. Lastly, there are more than 800 clinical annotations associated with 94 genes in BC pharmacogenomics.

genomics↗

bioRxiv2019

CompScore: boosting structure-based virtual screening performance by incorporating docking scoring functions components into consensus scoring

Consensus scoring has become a commonly used strategy within structure-based virtual screening (VS) workflows with improved performance compared to those based in a single scoring function. However, no research has been devoted to analyze the worth of docking scoring functions components in consensus scoring. We implemented and tested a method that incorporates docking scoring functions components into the setting of high performance VS workflows. This method uses genetic algorithms for finding the combination of scoring components that maximizes the VS enrichment for any target. Our methodology was validated using a dataset that contains ligands and decoys for 102 targets that has been widely used in VS validation studies. Results show that our approach outperforms other methods for all targets. It also boosts the initial enrichment performance of the traditional use of whole scoring functions in consensus scoring by an average of 45%. CompScore is freely available at: http://bioquimio.udla.edu.ec/compscore/

bioinformatics↗