Posters - Schedules

In this study, TextNetTopics was used to detect significant topics. Each topic is a set of words detected by the LDA approach. The Random forest then trained on the top k topics and tested on the test part of the data. Moreover, to improve the model's performance and deal with unbalanced test/validation data, we have suggested using the probability distribution to pick up the discriminating threshold that yields improved performance. The results showed that we could improve the performance by about 5% to 12% when applying the threshold tuning technique; the validation datasets' outcome indicates that our TextNetTopics is a stable tool.

Recent advances in high-throughput sequencing technologies provide the scientific community with efficient bioinformatics tools to profile human adaptive immune receptor repertoire via human adaptive immune receptor repertoire sequencing (AIRR-Seq). However, few studies have taken the ancestry information into account, and the open AIRR-Seq studies have unknown ancestry distribution. In this study, I examined the completeness of the immunogenetics database1 (IMGT). By leveraging the bioinformatics software, MiXCR2, I’m able to comprehensively examine the mismatches in different ancestry group samples’ read in the VDJ gene and evaluate the completeness of the immunogenetics database across diverse ancestry groups. Unveiling the ancestry distribution in TCR-Seq studies and the completeness of the immunogenetics database representing diverse populations could highlight the need to improve ancestry diversity in those underrepresented populations and guide future immunogenomics studies to improve ancestry availability and distribution.

Intercellular attachment between cells is potentially significant in COVID-19. However, the interactions among the responsible molecules have not been adequately uncovered. Limited understanding of how such molecules are regulated in COVID-19 causes incomplete signaling pathways. For example, β-actin (ACTB), ICAM-1 (ICAM1), and MOCCI (C15orf48) are not on the existing pathway databases, such as COVID-19 Disease Map or Signor 2.0. Therefore, this research aims to construct pathways associated with the above three genes of interest (GOIs). We propose a general framework for inferring gene networks from single-cell transcriptome data and six different knowledge bases and apply the framework to the GOIs at three-time points. Firstly, Markov random field retrieved differentially coexpressed genes specific to COVID-19 with spurious correlations deleted. Secondly, background knowledge validates the edges obtained from data. Lastly, pathway enrichment analysis using KEGG pathways discovers the nine GOIs-associated-pathways. These pathways suggest the immune response typical in COVID-19 and the possibility of membrane fusion or microtubule organizing center formation in COVID-19. In this manner, data-driven and knowledge-based approaches are harnessed into gene network inference for pathway construction. The results can contribute to repairing and completing the pathway databases, improving our current understanding of the COVID-19 mechanisms.

Motivation: Gene annotation is the problem of mapping proteins to their functions represented as Gene Ontology terms, typically inferred based on the primary sequences. Gene annotation is a multi-label multi-class classification problem which has generated growing interest for its uses in the characterization of millions of proteins with unknown functions. However, there is no standard GO dataset taking care of the best practices that can be used to properly benchmark the newly developed new machine learning models within the bioinformatics community. Thus, the significance of improvements for these models remains unclear.
Summary: The Gene Benchmarking database is the first effort to provide an easy to use and configurable hub for the learning and evaluation of gene annotation models. It provides easy access to preset datasets, and takes the non-trivial steps of preprocessing and filtering all data according to custom presets using a web-interface. The GO bench web application may also evaluate and display any trained model on leaderboards for annotation tasks.

In eukaryotic cells, the genome is organized into chromatin, a nucleoproteic structure that plays both a packing and functional role. Chromatin takes part in the regulation of gene expression, via different chromatin states modulating DNA accessibility by transcription factors and other chromatin binding factors. These functional states are characterized by different combinations of histone marks,
which correspond to modifications of the N-terminal regions of histones.

The main goals of our work are: 1. Find a method to quantify the variability of the epigenomic information and use it to dissect the epigenome’s role in the cellular differentiation, or the transition between a healthy and a cancerous state. 2. Explore approaches for the joint integration of epigenomic and transcriptomic mutual information, aiming at identifying regulatory modules and key drivers of the different cellular states.

To address our question, we have applied: matrix factorisation approaches such as Multi Factorial Analysis (MFA) and Independent Component Analysis (ICA); machine learning procedures, like Self organising maps, on a variety of epigenomic and transcriptomic datasets.

We have confirmed the validity of our approach, retrieved factors and/or signatures and we are now working on their biological interpretation.

Adverse drug-drug interaction (DDI) is a major concern to polypharmacy due to its unexpected adverse side effects and must be identified at an early stage of drug discovery and development. Many computational methods have been proposed for this purpose, but most require specific types of information, or they have less concern in interpretation on underlying genes. We propose a deep learning-based framework for DDI prediction with drug-induced gene expression signatures so that the model can provide the expression level of interpretability for DDIs. The model engineers dynamic drug features using a gating mechanism that mimics the co-administration effects by imposing attention to genes. Also, each side-effect is projected into a latent space through translating embedding. As a result, the model achieved an AUC of 0.889 and an AUPR of 0.915 in unseen interaction prediction, which is competitively very accurate and outperforms other state-of-the-art methods. Furthermore, it can predict potential DDIs with new compounds not used in training. In conclusion, using drug-induced gene expression signatures followed by gating and translating embedding can increase DDI prediction accuracy while providing model interpretability. The source code is available on GitHub (https://github.com/GIST-CSBL/DeSIDE-DDI)