Posters - Schedules

Human gut microbiome is complex and highly dynamic in nature. Dynamic patterns of the gut microbiome can capture more information than single point inference of the gut microbiome as it contains the temporal changes information of the gut microbiome. However, dynamic information of the human gut microbiome can be hard to be captured due to the complexity of obtaining the longitudinal data with large volume of missing values that in conjunction with heterogeneity may provide a challenge for the data analysis. We propose using an efficient hybrid deep learning architecture CNN-LSTM (convolutional neural network - Long Short-Term Memory), which is combined with self-knowledge distillation to create high accurate models to analyse the longitudinal gut microbiome to predict disease outcomes. Using our proposed models, we analysed the data sets from PROTECT (Predicting Response to Standardized Pediatric Colitis Therapy) study and DIABIMMUNE study. We showed the significant improvement in the area under the receiver operating characteristic curve (ROC-AUC) scores, achieving 0.889 and 0.798 on PROTECT study and DIABIMMUNE study respectively, compared to state-of-the-art temporal deep learning models. Our findings provide an effect artificial intelligence-based tool to predict disease outcomes using longitudinal gut microbiome profiles from collected patients.

According to the World Health Organization, Mycobacterium tuberculosis (Mtb) took the life of 1.5 million people in 2020, making it one of the actual deadliest worldwide diseases. Mexico is not an exception, in 2019 there were registered 19,000 of new cases and just under 2,000 deaths. Drug inefficiency is the leading cause of uncontrolled disease, caused by misuse of medications, incorrect prescriptions, or premature interruption of the treatment generating new immune strains. Existing treatments experiment with a decrement of efficiency due to drug-resistance strains, thus, it is key to detect characteristic signatures that indicate the presence of this immunity, this will avoid an upsurge in deaths and the emergence of more resistant strains. Bioinformatics and high-quality data have emerged as the principal support resources enabling the development of novelty tools that could help to understand these kinds of phenomena by analysing high-throughput omics data at different levels. We implemented FunTB. A novel functional networking python standalone tool, that allows comparing gene variations among groups of specific phenotypic features samples highlighting own and common alterations of each class, taking as input an MTBSeq file containing variant calling information and a series of lists of samples user wants to contrast, the pipeline provides as output a network preview image and a set of XML-based files, compatibles with specialized software as Cytoscape and Gephi for posterior processing. In this work we present the process of software implementation, explain its functioning and test it on real Mexican and international MTb databases, helping to contribute in the discovery of novel resistance signature candidates.

Currently prescribed antibiotics target predominantly the catalytic site of bacterial proteins. Moreover, they focus on wild-type proteins; however, bacteria adopt mutations to survive in their challenging environment. According to World Health Organization report (2022), resistant bacterial infections cause annually 4.95 million deaths. Therefore, there is an urgent need for identifiying alternative allosteric regions to combat with this life-threatening problem. Towards this end, we set out to investigate impact of high-resistance causing triple mutation (S385T + L389F + N526K) on dynamics of a prioritized resistant pathogen, H.influenzae, by computational techniques. We focused on FtsW-penicillin-binding protein 3 (PBP3) complex, which was shown to display resistance towards β-lactam antibiotics that prevent bacterial cell wall synthesis. We identified a reaction coordinate, an angle on the N-terminal periplasmic modulus (N-t) of PBP3 that displayed differences between the wild type and mutant protein. Interestingly, it adopted smaller values in the latter, which were associated with higher number of residues participating in the allosteric communication network on the pathway that connects the N-terminal module to the catalytic site. Moreover, the angle also impacted orientation of the transpeptidase domain with respect to the cytoplasmic membrane and the exposure of the catalytic site. In the wild type, the catalytic site was more exposed which presumably enabled accession of the site by antibiotics. We also picked up representative conformations of the wild-type and the mutant protein where we performed covalent-docking on the catalytic site using β-lactam antibiotics such as cefixime, ceftaroline, and amoxicillin. The mutant PBP3 displayed higher binding affinity than the wild type suggesting that it might hydrolyze the antibiotic more effectively. All these data suggest that the newly identified region can be targeted to modulate the activity of the resistant PBP3. Therefore, this study provides a new perspective for developing effective molecules to combat with resistant bacteria.