Tutorials

RSGDREAM 2022 Tutorial Program

Virtual Tutorials: (All times PST) Presented through the RSGDREAM conference platform

• Tutorial VT1: Interpreting deep neural network in genomics
• Tutorial VT3: GVViZ &PAS:  Integrated bioinformatics and mobile applications for gene-disease data annotation, expression analysis, and visualization for translational research

Virtual Presentation - Presented through the ISMB conference platform
Tuesday, November 8, 8:00 am - 12:00 pm PST

Organizer(s):
Yaron Orenstein, Ben-Gurion University
Ofir Yaish, Ben-Gurion University

Deep-learning applications in genomics have become instrumental in modeling molecular phenomena and enabling predictions to new molecules and DNA, RNA, and amino-acid sequences. Still, interpreting deep neural networks is a key open problem, especially in bioinformatics, where model interpretability is a requirement in any study to improve our understanding of the underlying molecular mechanism. In the tutorial, the participants will be acquainted with the state-of-the-art and most efficient techniques to interpret deep neural networks in genomics, both on the theoretical level and the practical level. They will understand the analytics behind different interpretation techniques, both classic and new, and their respective advantages and limitations. In addition, they will gain hands-on experience in running the most advanced techniques on a real bioinformatic application and will be able to apply them to their own research following their participation in the tutorial.

Learning Objectives for Tutorial:
The field of genomics has seen a flood of deep neural network applications in recent years. Deep neural networks have been outperforming many classic machine-learning techniques, and hence are being widely used in many bioinformatics applications. While their prediction performance is unprecedented, many researchers still see them as black-boxes, which limit our ability to understand and decipher biological phenomena on the molecular level. In this tutorial, we will cover the main techniques in interpretability of deep neural networks in genomics. We will start with interpretability techniques of classic methods, such as sequence logos, and show how they apply to one-dimensional sequence kernels in convolutional neural networks. We will move to new and highly efficient techniques based on gradient calculations, and present their performance in various tasks. We will conclude with a presentation of a motif finding algorithm to detect re-occurring local patterns in interpretability maps. Moreover, we will dedicate at least one hour to a hands-on exercise on state-of-the-art interpretability techniques. In the conclusion of the tutorial, the participants will be acquainted with the state-of-the-art and most efficient techniques to interpret deep neural networks in genomics, both on the theoretical level and the practical level. They will understand the analytics behind different interpretation techniques, both classic and new, and their respective advantages and limitations. In addition, they will gain a hands-on experience in running the most advanced techniques on a real bioinformatic application, and will be able to apply them to their own research following their participation in the tutorial.

Intended audience and level:
The level is similar to a graduate-level course, including the level of mathematical derivation and programming examples that will be presented through the tutorial. The course is indented to participants with a computational background and knowledge on and experience with applying deep neural networks to genomic data.

Maximum Participants: 100

- top -

Virtual Presentation - Presented through the ISMB conference platform
Tuesday, November 8, 10:00 am - 12:00 pm PST

Organizer(s):
Zeeshan Ahmed, Rutgers Institute for Health & Rutgers Biomedical and Health Sciences
Habiba Abdelhalim, Rutgers Institute for Health

Over the last few decades, genomics has led in changing our views on conducting biomedical research, studying diseases, and understanding diversity in our society across the human species. Investigating disease-causing genes can support finding the root causes of uncertainties in patient care. Although approaches that combine clinical and genomic information are becoming increasingly common, scientists and health care providers still are faced with the daunting challenge of identifying what genes may be relevant to the part of the body or biological system they are studying, and how variants may impact health in unique ways for each patient. Independent, and timely high-throughput next generation sequencing data analysis is still a challenge for non-computational biologists and geneticists. We are focused on helping researchers, medical practitioners, and pharmacists in having a broad view of genetic makeup that may be implicated in the likelihood of developing certain diseases. We emphasize that automated, user-friendly, and interactive genomics data analysis, visualization, and sharing should be an indispensable component of modern era, as it has potential to bridge the gap between algorithmic approaches and the cognitive skills of users and investigators.
In this tutorial, we present two bioinformatics applications (GVViZ and PAS), which are designed, developed, and freely available to support translational research and precision medicine. GVViZ is a user-friendly, cross-platform, and database application for RNA-seq-driven variable and complex gene-disease data annotation, and expression analysis with dynamic heat map visualization. It can assimilate patients’ transcriptomics data with public, proprietary, and our in-house developed gene-disease databases to query, easily explore, and access information on gene annotation and classified disease phenotypes with greater visibility and customization. Our gene-disease databases are globally accessible through PROMIS-APP-SUITE (PAS), and consists of authentic and actionable genes, SNPs, and classified diseases and drugs data collected from different clinical and genomics databases worldwide. PAS is a smart phone application, integrating ICD (9 and 10) and NDC codes, and sets of over 50,000 genes, over million SNPs, over 200,000 gene-disease combinations. PAS is designed to simplify navigation across the landscape of gene annotation resources by an efficient mobile record search engine, which is based on standardized genes and related diseases to help explore multi-purpose clinical and genomics concepts in meaningful ways. The performance of GVViZ and PAS have been tested and validated in-house with multiple experimental analyses.
The execution of GVViZ and PAS is based on a set of simple instructions that users without a computational background can follow to design and perform customized data acquisition and analysis. It utilizes processed RNA-seq data, and robustly visualize patterns and problems that may give insight into a patient’s genomic profile, unravel genetic predisposition, and uncover genetic basis of multiple disorders. Experts in clinics and researchers in life sciences can use GVViZ to visualize and interpret the transcriptomics data, making it a powerful tool to study the dynamics of gene expression and regulation. With the successful deployment in clinical settings, GVViZ and PAS has the potential to enable high-throughput correlations between patient diagnoses based on clinical and genomics data.

Learning Objectives:

• Genomics and transcriptomics, and of role RNA-seq driven data processing, gene expression analysis and interpretation.
• Understanding the importance of data visualization, and mapping and rendering large RNA-seq datasets.
• Learning using GVViZ platform for the automated gene-disease data annotation, expression analysis, and dynamic heat map visualization.
• Learning using PROMIS-APP-SUITE (PAS) to access and explore information about authentic genes (protein-coding and noncoding), mutations, gene-disease relationships, and classified diseases and drugs and their codes.
• Training to how to download, configure, use, and customize GVViZ and PAS applications, and model genomics databases.

Intended audience and level:
This tutorial will be aimed to the audience of any level (e.g., beginner, experience, advanced knowledge), mainly interested in learning about automated and integrated gene-disease data annotation, expression analysis, and visualization for translational research.

We will expect and equally appreciate presence of computational and non-computational scientists, bench scientists, bioinformaticians, biologist, geneticists, clinicians, and most importantly graduate and undergraduate students of life and medical sciences.

Maximum Participants: 50

- top -