Mini-symposium

Biology, and in particular omics (genomics, transcriptomics, regulomics, 3D genomics, meta-genomics, …), are generating very large amount of heterogenous data (big data) underlying very complex relations, for which traditional machine learning algorithms might fall to produce accurate results. Recent applications of deep learning have improved performance of most of traditional machine learning methods in omics, including protein binding prediction, cancer (sub-)type classification, omics data visualization, disease outcome prediction, precision medicine, genome-wide association studies, image classification, microscopy image denoising and processing. Moreover, deep learning was used to automatically extract relevant biological features without human knowledge, unlike machine learning which requires to preliminary define features for building models. In addition, deep learning has also enhanced the resolution of omics data, notably using deep convolutional neural networks, which can improve the power of downstream bio-statistical analyses.

In this mini-symposium, we will present an overview of deep learning applications to omics, in particular genome representations (Laurent Jacob), but also a review on regulatory sequence prediction (Raphaël Mourad), phenomics-trascriptomics data integration (Vera Pancaldi), prediction of G-quadruplexes (Elissar Nasseredine/Vincent Rocher), repeat transcription prediction (Mathys Grapotte), automatic histopathology diagnosis (Arnaud Abreu).

Detailed program

Third generation sequencing, examplified by Oxford Nanopore Technology, is becoming a mature high-impact technology. It has demonstrated great potential in solving the structure of complex genomes, such as polyploid genomes or genomes with a large number of repeated regions. It provides a better picture for studying transcriptomes and modeling alternative transcription. However, analyzing this data is still computationally challenging due to a high rate of sequencing errors (approximately 10%), the volume of the data and the intrinsic complexity of the biological objects involved.

The goal of this mini-symposium would be to take stock of the bioinformatics issues raised by the analysis of long reads and the recent contributions that have been proposed to solve those issues. Those contributions are based on methodological advances, such as asuccinct indexes, minimizers, seeds... We also want to provide practical guidelines to work with long reads. This includes the following topics:

• assembly of genomes and metagenomes;
• hybrid correction;
• self-correction;
• variant calling and structural variants;
• discovery and quantification of alternative transcripts.

Detailed program

Structural Bioinformatics as a discipline is particularly well-suited for the functional exploitation of OMICS data in the context of cellular networks and/or complex communities. Its main goal is to decipher sequence-structure-dynamics-function relationships within complex systems, embedded in their biological processes. This domain integrates experimental data produced by biophysical protocols, using efficient methods and algorithms contributed by applied mathematics and computer science, and benefits from the overwhelming capacities of supercomputers. JOBIM 2020 in Montpellier represents a unique opportunity to exchange ideas and views with the French Bioinformatics community at-large. The original content of the event, often under-estimated, will be focused this year to the latest challenges and developments addressing the 3D modeling of RNA structures and the comprehension of proteins/RNA interactions, and its correlation with biological functions based on the integration of sequencing data, leading to applications in protein design.

Detailed program