"Introduction to the knowledge transfer in translational biology"

Abstract: I will introduce the session speakers and the theme of the translational biology session. I will discuss the potential of applying the advanced knowledge transfer algorithms for biomedical applications.

Bio: Aidong Zhang is a William Wulf Faculty Fellow and Professor of Computer Science and Biomedical Engineering in the School of Engineering and Applied Sciences at University of Virginia. She is also affiliated with the Data Science Institute at University of Virginia. Her research interests focus on data mining, machine learning, bioinformatics and health informatics.

Keynote: "Learning how somatic mutability shapes cancer progression risk"

Abstract: Cancer has long been recognized as a disease of aberrant evolution, in which evolutionary selection and drift (or diversification) of cancer cell lineages lead to tumorigenesis and disease progression. The advent of large-scale genomic profiling has, however, dramatically revised our understanding of the details of that process. One particularly important observation has been a recognition of the broad range of hypermutability phenotypes that distinguish cancers or precancerous tissues from another in how they generate genetic or epigenetic diversity, which in turn shape a tissue’s somatic evolutionary landscape and its risk of disease progression. We sought to understand how variations in mechanisms of evolutionary diversification interact with selective forces and other factors to underlie cancer progression risk, using a machine learning approach to quantify the influence of features describing variation in mutational mechanisms relative to other potential factors driving disease progression risk. Our results suggest that a large portion of cancer progression risk is determined by variations in mutability phenotypes, a finding with potential implications for cancer treatment approaches and for the development of novel diagnostics of early cancer or precancer and progression outcomes such as recurrence, therapy resistance, or metastasis. In ongoing work, we seek to improve models of these processes and extend them to broader ranges of cancer types, progression processes, and mutation mechanisms

Bio: Dr. Schwartz received his PhD in Computer Science from MIT and performed postdoctoral work at the MIT Biology Department and as an Informatics Research Scientist at Celera Genomics Corp. In 2002, he joined the faculty of Carnegie Mellon University, where he is currently a Professor in CMU’s Department of Biological Sciences and Head of its Computational Biology Department. His research has covered a variety of problems in modeling, algorithms, and machine learning for computational genetics, genomics, and biophysics, with a current focus largely in computational cancer biology, clonal evolution, and somatic mutation mechanisms.

"Counting motifs on evolving network topologies"

Abstract: Biological networks of an organism show how different bio-chemical entities, such as enzymes or genes interact with each other to perform vital functions for that organism. Dr. Kahveci's lab is focusing on developing computational methods that will help in understanding the functions of large-scale biological networks. One of the fundamental and computationally challenging problems in analyzing biological networks is to identify and count the embeddings of a given motif. In this talk, we will discuss the challenges arising from the dynamic nature of biological networks from the point of view of the motif counting problem. Counting motifs is a computationally intensive problem, especially for large networks. Although existing methods can solve this problem for a given network and motif topology, they quickly become infeasible for evolving networks, as solving the motif counting problem every time the network topology goes through some change requires solving the baseline motif counting problem from scratch. In this study, we demonstrate our method which avoids the costly motif counting problem at subsequent states of the evolving network topology by incrementally updating this count based on the current and new states of the network. We show this for both edge dependent and independent motif count measures. We demonstrate that, with a negligible amount of book-keeping cost, we can maintain the motif count with close to perfect accuracy for a sequence of thousands of evolving network states, when it is impossible to do so using traditional methods.

Bio: Tamer Kahveci received his Ph.D. degree in Computer Science from University of California at Santa Barbara in 2004. He is currently a Professor and Associate Chair of Academic Affairs in the Computer and Information Science and Engineering Department at the University of Florida, serving as the Associate Chair of Academic affairs. Dr. Kahveci received the Ralph E. Powe Junior Faculty Enhancement award in 2006, CSB best paper award in 2008, the NSF Career award in 2009, the ACM-BCB (Bioinformatics and Computational Biology) best student paper award in 2010, ACM-BCB honorary best paper award in 2011, and BiCoB best paper award in 2018. His research focuses on bioinformatics. He has worked on indexing sequence and protein structure databases, sequence alignment and computational analysis of biological networks.

Dr. Kahveci has served as the PC co-chair of the ACM BCB conference in 2012 and 2017, the BioKDD workshop and the International Workshop on Robustness and Stability of Biological Systems and Computational Solutions in 2012, the Workshop on Epigenomics and Cell Function in 2013, and the Workshop on Computational Network Analysis, the Workshops Chair of the ACM-BCB conference in 2014, 2015, 2016, 2017, 2018, and 2019. He served as the Tutorials Chair of the ACM BCB and the IEEE BIBM conferences in 2015, and Workshop Chair in 2016. He is a member of the governing board of the ACM SIGBIO and the chair of the steering committee member of the ACM-BCB. He is a member of the editorial review board for of the journal International Journal of Knowledge Discovery in Bioinformatics (IJKDB). He was the lead guest editor of the Journal of Advances in Bioinformatics, special issue on "Computational analysis of biological networks" and associate editor in IEEE/ACM Transactions on Computational Biology and Bioinformatics. In addition to these, he has served on the program committees of numerous computational biology and database conferences.

"Decoding cancer cell maps to guide precision medicine"

Abstract: Most drugs entering clinical trials fail, often related to an incomplete understanding of the mechanisms governing drug response. Machine learning techniques hold immense promise for better drug response predictions, but most have not reached clinical practice due to their lack of interpretability and their focus on monotherapies. We address these challenges by developing DrugCell, an interpretable deep learning model of human cancer cells trained on the responses of 1,235 tumor cell lines to 684 drugs. Tumor genotypes induce states on cellular subsystems which are integrated with drug structure to predict response to therapy and, simultaneously, learn biological mechanisms underlying the drug response. DrugCell predictions are accurate in cell lines and also stratify clinical outcomes. Analysis of DrugCell mechanisms leads directly to design of synergistic drug combinations, which we validate systematically by combinatorial CRISPR, drug-drug screening in vitro, and patient-derived xenografts. DrugCell provides a blueprint for constructing interpretable models for predictive medicine.

Bio: Jisoo Park is a computational biologist in oncology at the Genomics Institute of the Novartis Research Foundation whose research is focused on identifying target/mechanism of actions/biomarker of cancer treatments. Prior to her current position, she earned her Ph.D. in Computer Science at Tufts University after publishing multiple network biology studies aimed at identifying disease-mediating pathways. She then continued her academic career as a postdoc at UCSD where she has developed interpretable artificial intelligent (AI) models of cancer cells and performed differential interaction analyses of cancer PPI network with an aim of discovering novel cancer therapeutic interventions.

"Better inference through chemistry: knowledge-guided inference of biomolecular kinetics"

Abstract: Recent advances in time-series analysis have enabled fitting of biomolecular kinetics using empirical state models. Here, we discuss how the laws of physical chemistry can be used to structure inference problems, yielding richer and more informative models.

Bio: Peter Kasson is an Associate Professor of Molecular Physiology and of Biomedical Engineering at the University of Virginia and a Wallenberg Academy Fellow at Uppsala University. His research develops and applies physical and computational tools to understand complex biology, particularly applied to emerging pathogens. Some areas of focus have been developing tools for ensemble simulation and analyzing single-event statistics in enveloped viral infection.

"Predicting evolution with protein language models"

Abstract: The degree to which evolution is predictable is a fundamental question in biology. By learning the "grammar" that governs how amino acids can appear together to form a protein sequence, we can begin to predict the evolution of proteins. First, we use language models to predict mutations that enable viral proteins to escape from immune pressure. We identify escape mutations as those that preserve viral infectivity but cause a virus to look different to the immune system, akin to word changes that preserve a sentence’s grammaticality but change its meaning, and validate our predictions using proteins from influenza, HIV, and SARS-CoV-2. We then extend these ideas beyond viral evolution, showing how language models can recover complex evolutionary dynamics of protein evolution, enabling us to predict the directionality of evolution over vastly different timescales, from viral proteins evolving over years to eukaryotic proteins evolving over geologic eons.

Bio: Brian Hie is a Stanford Science Fellow in the Stanford University School of Medicine, where he is using algorithms to better understand host-pathogen interactions. As a doctoral student in electrical engineering and computer science at MIT, he developed novel applications of machine learning to biology and achieved important insights into viral mutation. He has also worked on machine learning for health-related applications at Google X and Illumina..