We are all familiar with some of the components of biological systems – DNA, proteins, cells, organs, individuals – but understanding biological systems involves more than just cataloging its component parts. It is critical to understand the many interactions of these parts within systems, and how these systems give rise to biological functions and responses and determine states of health and disease. The National Resource for Network Biology provides the scientific community with a broad platform of computational tools for the study of biological networks and for incorporating network knowledge in biomedical research.
We are planning new technology research projects focused on three complementary themes for which NRNB and its collaborators are well-positioned to catalyze major change (below in bold). These three themes serve as the major organizing principles underlying each of the three project areas. All three projects capitalize on the rapidly increasing accumulation of molecular network data in mammals, and the concomitant growth of new ‘omics profiling technologies which creates remarkable new opportunities for using network-based approaches to understand disease states on a genome-scale in human individuals.
1. Generating Differential and Dynamic Networks. New mass spectrometry technology is making it possible to capture comprehensive changes in protein expression and phosphorylation at lower cost, higher speed and higher spatial resolution, which makes it possible to measure differential network expression information in clinical samples and across tissues and time points. Single-cell genomics, including single-cell RNA-seq, now achieves high resolution measurements of transcriptional state on a per cell basis over multiple time points, enabling differential and dynamic network expression information to be measured or inferred. In this area, we develop new computational technologies that take advantage of these qualitatively new data types to better understand how networks function in differential biological conditions, such as cell types and disease states, and to infer whole-cell dynamic network models. The goals of these technologies are to [1] capture the molecular information flow from targeted perturbations to downstream cellular responses; [2] functionally characterize mechanisms defining individual cell types and their positions in developmental lineages; and [3] capture and visualize differential and dynamic changes in protein interactions across biological contexts, such as disease versus normal tissue types.
2. Modeling Multi-scale Network Architecture. In this project, we advance methods to transition Network Biology from flat diagrams of nodes and edges towards multi-scale models of biological systems. Although current network models and layouts provide a useful summary of an interaction dataset, these models and visualizations do not capture the exquisite multi-scale hierarchy of modular components and subcomponents that are evident in many biological systems – from amino acids to proteins to protein complexes to biological processes to organelles, cells, and tissues. Previously, we demonstrated that detailed hierarchical information can be systematically revealed by biological network modelling. This discovery enabled us to reconstruct and greatly extend the Gene Ontology (GO) hierarchy, yielding a GO built directly from protein networks (Data-Driven Ontologies) rather than literature. Here we seek to significantly increase the resolution and scalability of algorithms for detection of hierarchical network structure [1] and the ability to integrate (embed) many different lines of network evidence in discovering such structure [2]. We also aim to broaden and generalize the concepts of hierarchical network analysis to study biological structure at the larger scales of cell populations and tissues [3].
3. Network-guided Machine Learning for Biomedicine. Powerful machine learning technologies, such as the recent advances in deep learning, promise to revolutionize our ability to make predictions in biology and medicine and could one day replace people in tasks such as image analysis and prediction of patient drug response. However, machine learning models are typically black boxes that are challenging to use to gain the mechanistic understanding required for us to control and fix biological systems for industrial and medical applications. How can we reconcile these approaches to gain both the predictive power of machine learning and the interpretability of our mechanistic models of biology? Here we explore a series of complementary and innovative approaches to this question based on integrating machine learning models with biological networks. Specifically, we aim to use networks to: [1] Guide the transfer of predictive models of drug response from model systems to patients; [2] Apply machine learning models to genotype-phenotype prediction in genome-wide association studies; and [3] Use machine learning for patient diagnosis and clinical trial selection in precision medicine applications.
To serve rapidly growing community of network biology researchers, the NRNB runs a very active program of Community Engagement. This program includes Training, Dissemination, and Service components (below in bold). While the community of researchers benefiting from network biology is already substantial, one suspects that many other investigators (basic, clinical, and pharmaceutical) as of yet have little knowledge of recent developments in network biology and would benefit greatly from community engagement centered on NRNB technologies. As our engagement efforts continue, the network biology community could potentially become a good deal larger.
Training. We regularly teach application of NRNB technologies in multiple formats and in diverse research areas. Each year we lead over a dozen workshops and seminars across multiple cities and countries. We have matched over 100 students with network biology mentors as part of Google’s Summer of Code (GSoC) program, for diverse set of successfully completed projects. We have also offered a full university course in Network Biomedicine at UCSD which has been very well received by the students. We are planning to expand our education and training platforms to be more extensible and modular, to scale to a substantially larger community of trainees, and to recruit and empower community members to become trainers themselves.
Dissemination. We have established a collection of interlinked websites, social media feeds and community forums for the dissemination of NRNB technologies and to provide general information about the field of Network Biology. Regular activities include prolific posts on social media of late-breaking network biology publications and results and the establishment of a network biology helpdesk, a professional LinkedIn group, and a portfolio of tutorial YouTube videos. We launched the NetBio Community of Special Interest which has recently been promoted to a central track of the Intelligent Systems for Molecular Biology (ISMB) annual international conference. Our NRNB software ecosystem provides a major means of dissemination for bioinformatic tools, with connections to platforms like Cytoscape and Jupyter/Python/R.
Service. We respond to 60-80 requests for collaboration and service in any given year. These service collaborations range from assisting users with the application of methods and tools for network analysis to consulting on the development of Cytoscape apps. Service activities naturally provide access to, and facilitate adoption of, many center-developed tools.