Invited Speakers

Title: Computer Vision Opportunities for Shape Analysis from EM Connectomics Data
Start Time: 9:15 AM PDT
Speaker: Sharmishtaa Seshamani, Ph.D., Allen Institute for Brain Science
Abstract: The field of connectomics aims to map the morphology and synaptic connectivity of all neurons in a brain region. Understanding the detailed wiring patterns of the brain is of fundamental interest not only in neurobiology, but also for novel biological-inspired architectures in artificial intelligence. Electron Microscopy (EM) makes it possible to densely label diverse biological objects across more than five orders of magnitude in size, from synapses and cellular organelles tens or hundreds of nanometers wide to neuronal arbors more than a millimeter long in datasets containing tens of thousands of neurons. The current largest EM image datasets are on the petabyte scale, including the entire fruit fly brain and a cubic millimeter of a mouse visual cortex.
    The ability to assemble and analyze such large scale image data relies heavily on advances in computer vision, such as image registration to assemble 3D image volumes, segmentation to extract neuronal morphology and shape analysis to describe neurons of different cell types. We present our efforts towards building such a dataset which includes tissue processing, volume assembly and information extraction. We have recently released a public dataset of the mouse visual cortex (layer 2/3) and are currently actively pursuing several types of analysis of this data for studying connectivity and morphological cell typing.
Bio: Dr. Sharmishtaa Seshamani is part of the Cell Types team at the Allen Institute for Brain Science. Her current research focuses on building a data processing pipeline to image cells in mouse and human brain, to elucidate the structural complexity of cell morphology and ultrastructural connectivity of neural circuits. She received her Ph.D. in Computer Science from Johns Hopkins University, where she focused on developing meta-registration and mosaicking techniques for endoscopic imaging. Prior to joining the Allen Institute, Dr. Seshamani was a Senior Fellow in the Biomedical Image Computing Group at the University of Washington. There, she developed a pipeline for fetal fMRI analysis which included algorithms for MRI artifact correction, motion estimation and reconstruction. Currently Dr. Sheshamani develops tools for linear and elastic registration and 3D image reconstruction optimized for large datasets, including electron microscopy (EM) data as part of the MICrONS project: https://microns-explorer.org/. Her interests include applications of machine learning techniques for automatic parameter selection and connectivity data analysis.
Title:Immunological Synapse Quality Predicts the Efficacy/Toxicity of CAR Immunotherapy
Start Time:10 AM PDT
Speaker: Dongfang Liu, Associate Professor, Rutgers University- New Jersey Medical School.
Abstract: Human immune system consists of a network of billions of independent, self-organized cells that interact with each other to form a highest intelligence. Artificial intelligence (AI) is being used in immunological research. Machine learning (ML) is an AI tool that can process huge data and generate models for immune system and immunological research. One of the most exciting, recent breakthroughs in immunological research is cancer immunotherapy. Specifically, one such therapy involves engineering immune cells to express chimeric antigen receptors (CAR), which combine tumor antigen specificity with immune cell activation in a single receptor. Recent clinical trials testing cancer immunotherapies have shown promising results for treating various cancers. The adoptive transfer of these CAR-modified immune cells (especially T-cells, CAR T) into patients has shown remarkable success in treating multiple refractory blood cancers. To improve their efficacy and to expand their applicability to other cancer types, scientists are optimizing different CARs with different modifications. However, predicting and ranking the efficacy of different CAR T cell products with identical antigen specificity, selection of responders with an identical CAR T construct, and identification of the optimal CAR for a translation researcher to further develop potential clinical application are limited by the current, time-consuming, costly, and labor-intensive conventional tools used to evaluate efficacy. Particularly, T cell efficacy is not only controlled by the specificity and avidity of the tumor antigen and T cell interaction, but also it depends on a collective process, involving multiple adhesion and regulatory molecules, spatially organized at the T cell immunological synapse (IS). Optimal function of cytotoxic lymphocytes depends on IS quality. Recognizing the inadequacy of conventional tools and importance of IS in T cell functions, we investigate a new strategy for assessing CAR T efficacy. Previous studies in our lab show evidence that: 1) CAR T-cell immunological synapse (IS) quality (structure, function, and signaling) varies between CAR T-cells, 2) CAR co-stimulatory endodomains influence IS quality, 3) CAR T IS quality correlates with antitumor activity both in vitro and in vivo, and 4) CAR IS quality assay developed here can distinguish a responder from non-responders. However, current IS quality image data analysis was quantified manually, which is time-consuming and labor-intensive with low accuracy. Here we develop a machine learning method to quantify thousands of synapse images with enhanced accuracy. The ML-based, automated algorithms to quantify CAR T IS can develop fast and easy tools to predict CAR T cell efficacy, which provides guidelines for designing and optimizing CARs for clinical cell therapy.
Bio: Dongfang Liu, PhD recently joined the Department of Pathology, Immunology & Laboratory Medicine and the Center for Immunity and Inflammation as an Associate Professor. In 2012, Dr. Liu was recruited to Baylor College of Medicine as a tenure-track Assistant Professor in the Department of Pediatrics and Department of Pathology & Immunology, before joining Houston Methodist Hospital (a teaching hospital affiliated with Weill Cornell Medical College) as an assistant professor in 2015. In 2018, Dr. Liu was promoted to an Associate Professor in Houston Methodist Research Institute. Dr. Liu did his postdoctoral training on natural killer (NK) cells at the National Institute of Allergy and Infectious Diseases (NIAID) in National Institutes of Health (NIH) from 2005 to 2011. After completing the postdoctoral training, he joined Ragon Institute of MGH, MIT and Harvard in 2011 as a senior research scientist, where he worked on HIV-specific CTL dysfunction with a focus on PD-1 in HIV- specific CTLs. His current research is primarily focused on the immunobiology of chimeric antigen receptor (CAR) T and NK cells, immunoreceptors, CAR immunotherapy, and HIV-specific CTLs in chronic HIV and its related malignancies, with a focus on immunological synapse biology and its clinical applications. His research is supported by several NIH grants, including an R01 and three R21 grants.
Title:Algorithms and Tools to Analyze Multi-Modal Brain Imaging Data
Start Time:1:15 PM PDT
Speaker: Kunal Ghosh, Ph.D., Inscopix.
Abstract: Decoding the brain is a grand challenge that will require innovations in both brain recording tools and in algorithms for extraction of signals from brain data. Traditional electrode-based recording tools do not have the cell type-specificity or the scale to capture the individual activity of a large number of genetically-targeted neurons, key to mapping the patterns of neural activity that give rise to behavior. Fluorescence microscopy, a workhorse tool for the life sciences, has emerged as a powerful modality to record brain activity with cellular resolution, cell-type specificity and at scale. Recent advances in head-mounted miniaturized fluorescence microscopes, or "miniscopes", now enable imaging activity from thousands of neurons during behavior and longitudinally over time, providing neuroscientists with unprecedented data sets of the brain circuits implicated in learning, memory formation, sensory processes, and in brain disease.
    These new data sets are rich in information about how the brain works, and how it does not, but are of little use if the information cannot be efficiently, reliably and accurately extracted. In this talk we will present methods to extract electricity from pixels, and methods and tools to analyze large brain imaging data. In the first part of the talk we will show how individual time traces of the activity of each neuron imaged can be extracted from terabytes of miniscope brain data, and in the second part of the talk we will present MIRA, a Multimodality Image Registration and Analysis toolbox, that enables integrating and co-registering functional miniscope neural imaging data with high-resolution microscopy data to help researchers map the brain.
Bio: Kunal Ghosh is founder and CEO of Inscopix, Inc., a Palo Alto-based startup developing a platform for mapping brain circuits. Kunal founded Inscopix out of a research project at Stanford University that resulted in the invention of a miniature, integrated microscope for in vivo brain imaging. The invention is the centerpiece of Inscopix's core brain circuit mapping products which today have already advanced fundamental knowledge of the brain circuits underpinning brain function and behavior. Inscopix's platform also has the potential of shaping the future of neuro-therapeutic discovery, enabling the development of entirely new in vivo circuit-based assays for diseases such as Parkinson's, epilepsy, and depression. Kunal is a passionate advocate of the role of business as a force for good in society, and especially the role of entrepreneurial ventures in catalyzing a technologyfueled revolution in brain science and mental health. He is a frequent speaker on these themes at scientific meetings, industry conferences, and top Business schools, and serves on the World Economic Forum's Council on Neurotechnologies and Brain Science. Kunal holds a BSE in Electrical Engineering from the University of Pennsylvania, a BS from the Wharton School at the University of Pennsylvania, and an MS and PhD in Electrical Engineering from Stanford University.
Title:Creating a Landscape of Cellular Signatures through Quantitative Live Cell Imaging
Start Time:3:15 PM PDT
Speaker: Winfried Wiegraebe, Ph.D., Director of Microscopy, Allen Institute for Cell Science.
Abstract: To build predictive models of the human cell, we develop tools to create quantitative and reproducible data sets. Human induced pluripotent stem cells (hiPSC) are healthy diploid cells which can differentiate into almost any cell type of the human body. They have a normal karyotype unlike many currently used cell lines. We created stable cell lines that express green fluorescence protein (eGFP) fused to proteins at their endogenous locus. The fluorescence intensity of these fusion proteins is proportional to their abundance.
    We established specialized imaging pipelines based on spinning disk microscopes optimized for throughput and point-scanning confocal microscopes with detectors that achieve an optical resolution beyond the classical diffraction limit. We are adding lattice light sheet microscopy to our toolbox for fast and long-term time lapse imaging. To ensure reproducibility, we developed software to automate imaging pipelines.
    We will present a workflow to calibrate the intensity between different microscope modalities and to estimate the absolute concentration of a given protein. We created a dilution series of eGFP solutions in the physiological relevant range from nM to M. We measured the concentrations with fluorescence correlation spectroscopy (FCS). Then we imaged the eGFP solutions with the same microscopes under identical settings as used for cell imaging. We created a concentration/intensity calibration curve that we applied to microscope images of cells to obtain the concentration of eGFP labeled protein. For most cell lines we inserted eGFP only into one allele. Quantitative Western Blots allowed us to estimate the ratio between labeled and unlabeled proteins in the cells. Using this ratio, we could estimate the total concentration of a given protein within life cells (labeled and un-labeled protein).
    We developed a toolbox to segment cellular structures. The extracted features create a landscape of cellular signatures that can be explored through our Cell Feature Explorer and are used as input for quantitative modelling. We believe in open science, and thus all of our cell lines, data sets, and tools are available at https://www.allencell.org.
Bio: Dr. Wiegrabe is the Director of Microscopy & Image Analysis at the Allen Institute. All his professional life, Winfried has used, programmed, and built microscopes. He worked in industry and academia, in Germany and the US. Winfried received his diploma in Biophysics and Machine Tools and Industrial Management from the Technical University in Munich, Germany. For his diploma and Ph.D. thesis he joined the Department of Molecular Structural Biology headed by Wolfgang Baumeister at the Max-Planck Institute for Biochemistry in Martinsried, Germany. Under the mentorship of Reinhard ‘Guckus’ Guckenberger, Winfried investigated hydrated bacterial surface proteins with scanning tunneling microscopy (STM). He developed an atomic force microscope (AFM) to measure their conductivity. He complemented this data with measurements of their local elasticity and friction. For a short post-doc, he joined the group of Eberhard Unger at the Institute for Molecular Biology in Jena, Germany. He applied low current STM to hydrated microtubules. Being in Jena, where Carl Zeiss founded his microscopy company and Ernst Abbe worked out the theory of the light microscope and its resolution limit, Winfried headed a R&D team at Carl Zeiss to develop a product based on the discoveries of Eric Betzig, the scanning near field optical microscope (SNOM) to overcome this limit. Later, he became world-wide product manager for fluorescence correlation microscopy (FCS), a method that can measure the concentration, diffusion properties, and molecular interactions of proteins within living cells. In 2001, Winfried joined the Carl Zeiss group in the USA and helped customers to be successful users of confocal and two-photon microscopes (NLO), as well as FCS. In 2005, Winfried moved back to academia and joined the Stowers Institute for Medical Research in Kansas City, Missouri, USA. He created the group for Advanced Instrumentation and Physics and became later the Head of the Stowers Microscopy Center. In this capacity he supported and built with his team a large variety of microscope techniques, from laser micro-dissection to super-resolution techniques and light sheet microscopy. He developed technology to automatically perform FCS measurements on 4000 different proteins in yeast.