Harnessing Big Data for Precision Medicine and Healthcare

Introduction to Future Lab

A big data approach for personalised infection medicine

The immune function at an individual level is highly variable. It is therefore important to capture the impact of genetics and biological molecules on the immune function to better understand the immune system as a whole. In my talk, I will describe novel strategies to study these factors by simultaneously modeling information from genome, transcriptome, proteome, metabolome and environmental profiles. The results from such a study will reveal insights into the mechanisms driving the immune response to pathogens and provide mathematical models for predicting individual variation in immune functions, a crucial step towards personalized prevention/treatment of infectious diseases.

Yang Li

Multi Object Tracking for cells, microorganisms and human motion analysis

In this talk I will summarise recent results on multi object tracking using the tracking-by-detection paradigm. The challenge is to use detections (e.g. bounding boxes of objects in images) as input and the goal is to generate consistent trajectories over the entire sequence. This task can be cast as a network flow graph, which can be globally optimised using a linear programming (LP) formulation. I will present selected basics about (integer) linear programming, some simple discrete optimization tasks (e.g. graph matching), their respective LP formulations and finally explain graph based tracking and its optimisation. The talk continues with some applications, recent results e.g. on the MOTA benchmark and concludes with future research strands. Recent papers: Andrea Hornakova*, Roberto Henschel*, Bodo Rosenhahn, Paul Swoboda, (* equal contribution) Lifted Disjoint Paths with Application in Multiple Object Tracking Proceedings of the 37th International Conference on Machine Learning (ICML), July 2020 Roberto Henschel, Timo von Marcard, Rosenhahn Bodo Simultaneous Identification and Tracking of Multiple People using Video and IMUs Computer Vision and Pattern Recognition Workshops (CVPRW), June 2019 Timo von Marcard, Roberto Henschel, Michael J. Black, Bodo Rosenhahn, Gerard Pons-Moll Recovering Accurate 3D Human Pose in The Wild Using IMUs and a Moving Camera European Conference on Computer Vision, September 2018 Weblink: https://motchallenge.net

Prof. Dr.-Ing. Bodo Rosenhahn

A deep generative model for molecular graphs

Deep generative models have been praised for their ability to learn smooth latent representation of images, text, and audio, which can then be used to generate new, plausible data. However, current generative models are unable to work with molecular graphs due to their unique characteristics—their underlying structure is not Euclidean or grid-like, they remain isomorphic under permutation of the nodes labels, and they come with a different number of nodes and edges. In our work we propose a variational autoencoder for molecular graphs, whose encoder and decoder are specially designed to account for the above properties by means of several technical innovations. We further develop a gradient-based algorithm to optimize the decoder of our model so that it learns to generate molecules that maximize the value of a certain property of interest and, given a molecule of interest, it is able to optimize the spatial configuration of its atoms for greater stability.

Bidisha Samanta

Learning Under Algorithmic Triage

Under algorithmic triage, a machine learning model does not predict all instances but instead defers some of them to human experts. The motivation that underpins learning under algorithmic triage is the observation that, while there are high-stake tasks where machine learning models have matched, or even surpassed, the average performance of human experts, they are still less accurate than human experts on some instances, where they make far more errors than average. The main promise is that, by working together, human experts and machine learning models are likely to achieve a considerably better performance than each of them would achieve on their own. In this talk, I will present several algorithms to learn under algorithmic triage that we have developed in recent years, discuss their theoretical properties, and present a variety of experimental results demonstrating their potential in improving medical diagnosis, content moderation and scientific discovery.

Manuel Gomez Rodriguez

Big data and mental health - moving forward to precision psychiatry

One of the major obstacles in today’s psychiatric research is the obvious mismatch between the diagnostic categories stemming mainly from the end of the 19th century and modern neurobiological concepts of normal and disrupted brain function, leading to a lag in the development of new and more effective therapies/therapeutics. One main research goal of my group is the use of (epi-)genetic markers to detect and categorize biologically distinct sub-groups of psychiatric disorders, using therapy response as a primary phenotype. We were able to establish DNA methylation markers indicating an elevated risk for non-response to standard mono-aminergic antidepressants (EU and US patent granted) or for non-response under electro-convulsive therapy (ECT). Further DNA methylation markers that predict positive response to ECT and specific psychotherapeutic treatments are currently being tested. Large data sets derived from the different branches of basic or clinical research contain a wealth of information. To access and implement this information, new methods of data analysis are needed. Pattern recognition based on artificial intelligence/neuronal networks is a feasible approach to tackle these research questions. During recent years, we have gained expertise in the use of these “big-data” methods for the integrative analysis of molecular and clinical data. The use of self-learning algorithms not only helps to discover new and unexpected relationships between molecular and clinical data but also fosters the development of diagnostic and treatment algorithms in an iterative and evolutionary way (plan-do-check-act (PDCA)-cycle integrating patient’s care and research goals), paving the way for precision psychiatry.

Helge Frieling

Developing Smartphone Keyboard Interaction based Emotion Detection System

Keyboard interactions on communication applications, like WhatsApp, FB messenger can induce emotional exchanges. Moreover, monitoring keyboard interaction is unobtrusive and does not have high resource implications. As a result, exploring different types of keyboard interaction patterns (such as typing speed, touch pressure, error rate, special characters usage) can reveal emotion cues and aid in developing non-intrusive, resource- friendly emotion tracking applications. Additionally, the effects of emotion in one session of communication can persist across sessions, which if modeled jointly with keyboard interaction patterns, can further improve the emotion detection performance. We, in this work, discuss the design, development, and implementation of an Android application, TouchSense, which leverages keyboard interaction characteristics and implements a machine learning model to infer multiple emotion states (happy, sad, stressed, relaxed). We also discuss the approaches of automatically learning the keyboard interaction representation and adopting a Multi-task Learning strategy so that the keyboard interaction similarity among users can be leveraged for a superior performance.

Surjya Ghosh

Multimodal sensor machine learning for mental health

Digital phenotyping and machine learning technologies have shown a potential to measure objective behavioral and physiological markers, provide risk assessment for people who might have a high risk of poor health and wellbeing, and help better decisions or behavioral changes to support health and wellbeing. I will introduce a series of studies, algorithms, and systems we have developed for measuring, predicting, and supporting personalized health and wellbeing. I will also discuss challenges, learned lessons, and potential future directions in health and wellbeing research.

Akane Sano

Empathy in peer-to-peer mental health support

Access to mental health care is a global problem for hundreds of millions of people. Online mental health support may be able to help mitigate this global challenge. However, while mental health platforms attract millions of people, we lack a comprehensive understanding of how conversations on these platforms could be most effective in supporting those in need. Specifically, I will describe how we can measure empathy in mental health peer support and how we can give feedback in order to empower peer supporters to increase expressed levels of empathy, using large-scale neural transformer architectures and reinforcement learning approaches. I will also share learnings from applying these methods to a large corpus of support conversations.

Tim Althoff